WO2019235018A1 - Control system, control method, learning device, control device, learning method, and learning program - Google Patents

Control system, control method, learning device, control device, learning method, and learning program Download PDF

Info

Publication number
WO2019235018A1
WO2019235018A1 PCT/JP2019/010189 JP2019010189W WO2019235018A1 WO 2019235018 A1 WO2019235018 A1 WO 2019235018A1 JP 2019010189 W JP2019010189 W JP 2019010189W WO 2019235018 A1 WO2019235018 A1 WO 2019235018A1
Authority
WO
WIPO (PCT)
Prior art keywords
value
data
command value
learning
target device
Prior art date
Application number
PCT/JP2019/010189
Other languages
French (fr)
Japanese (ja)
Inventor
泰明 阿部
勇樹 上山
高史 藤井
和彦 今竹
Original Assignee
オムロン株式会社
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by オムロン株式会社 filed Critical オムロン株式会社
Priority to US17/052,523 priority Critical patent/US11681261B2/en
Priority to CN201980030426.2A priority patent/CN112105994B/en
Priority to EP19814529.4A priority patent/EP3805875B1/en
Publication of WO2019235018A1 publication Critical patent/WO2019235018A1/en

Links

Images

Classifications

    • G—PHYSICS
    • G05—CONTROLLING; REGULATING
    • G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
    • G05B13/00—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion
    • G05B13/02—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric
    • G05B13/0265—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric the criterion being a learning criterion
    • G—PHYSICS
    • G05—CONTROLLING; REGULATING
    • G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
    • G05B13/00—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion
    • G05B13/02—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric
    • G05B13/04—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators
    • G05B13/041—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators in which a variable is automatically adjusted to optimise the performance
    • G—PHYSICS
    • G05—CONTROLLING; REGULATING
    • G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
    • G05B13/00—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion
    • G05B13/02—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric
    • G05B13/04—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators
    • G05B13/048—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators using a predictor
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N5/00—Computing arrangements using knowledge-based models
    • G06N5/01—Dynamic search techniques; Heuristics; Dynamic trees; Branch-and-bound
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G06N20/10—Machine learning using kernel methods, e.g. support vector machines [SVM]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods

Definitions

  • the present invention relates to a control system, a control method, a learning device, a control device, a learning method, and a learning program.
  • Patent Document 1 proposes a control system that has a plurality of models to be controlled and calculates a predicted value of a controlled variable using any one of the models. Specifically, this control system selects a model used for prediction calculation according to the external environment, and determines an operation amount using the selected model. Thereby, the prediction control of the target apparatus according to the external environment can be realized.
  • the prediction model is constructed using learning data collected in advance. Therefore, according to this prediction model, it is possible to appropriately determine the command value for the target device in the case that is the same or similar to the situation that appears in the learning data, but in the unknown case, the command value is appropriately set. There is a possibility that it cannot be determined. In other words, when the operation of the target device is controlled in an unknown case, this prediction model may output an inappropriate command value such as a value that exceeds the movable range or a value that causes a failure. Therefore, when the prediction model is used, a constraint condition (threshold value) for limiting the range of the command value is provided in order to ensure the safety of the operation of the target device. For example, in the control system proposed in Patent Document 1, the optimum operation amount is determined from the predicted value of the control amount predicted using the model under the preset constraint conditions.
  • the constraint condition (threshold value) is basically set in advance by a user who uses the control system. At this time, considering the safety of the operation of the target device excessively, there is a possibility that the constraint condition is set so that the allowable range of the command value is narrower than the range satisfying the safety. If the constraint condition is set in this manner, the command value determined by the prediction model is accepted as the command value used for controlling the target device because the command value determined by the prediction model does not satisfy the constraint condition even though the safety is satisfied. Therefore, the predictive control cannot be properly performed. In other words, a control system that uses preset constraint conditions can ensure the safety of the operation of the target device, but may not be able to fully demonstrate the performance of the prediction model. The inventors have found out.
  • the present invention has been made in view of such a situation, and an object thereof is to provide a technique for performing predictive control capable of sufficiently exhibiting the performance of a predictive model.
  • the present invention adopts the following configuration in order to solve the above-described problems.
  • the control system is adapted to the first data related to the factor that determines the operation of the target device and the command value to the target device, which is indicated by the first data.
  • the learning data acquisition unit that acquires a plurality of learning data sets each configured by a combination of second data related to the command value and each of the acquired learning data sets are input the first data
  • a learning processing unit that constructs a prediction model so as to output a value corresponding to the second data, and a numerical value range that the command value can take from the distribution of the second data in the acquired plurality of learning data sets.
  • An estimation unit for estimation and a first allowable range defined by a first threshold set in advance for the command value to the target device are expanded.
  • a threshold value determination unit that determines a second threshold value for the command value to the target device, and an input data acquisition unit that acquires input data related to the factor in the operation phase; , By inputting the acquired input data to the prediction model, an output value is acquired from the prediction model, and a second allowable range defined by the second threshold value determined based on the acquired output value
  • a prediction calculation unit that determines a command value for the target device, and an operation control unit that controls the operation of the target device based on the determined command value.
  • the control system according to the configuration estimates a numerical range that can be taken by the command value to the target device from the distribution of the second data in the learning data set used for constructing the prediction model. And the control system which concerns on the said structure is the 1st with respect to the said command value based on the estimated numerical range so that the 1st tolerance
  • the control system according to this configuration uses the second allowable range defined by the second threshold value as a restriction condition for the command value. That is, the control system according to the configuration determines the command value for the target device within the second allowable range defined by the second threshold value in the operation phase using the prediction model.
  • the second allowable range set to expand the first allowable range is used as a constraint condition.
  • permits the command value used for control of an object apparatus can be expanded. That is, a part of the command value that is rejected when the first allowable range is used as a constraint condition can be used for controlling the target device.
  • each learning data set can be collected so as to realize control of operation suitable for a specific case, the command value specified based on the second data in each learning data set Accordingly, the operation of the target device can be safely controlled.
  • the second threshold value defining the second allowable range is determined so as to ensure the safety of the operation of the target device. can do. Therefore, according to the control system which concerns on the said structure, the predictive control which can fully exhibit the performance of a prediction model can be implemented, ensuring the safety
  • the “target device” may include all types of devices that can be controlled, and may include, for example, a production device configured to produce a product from a workpiece.
  • the “prediction model” is not particularly limited as long as it is a model that can predict a command value to the production apparatus at a time point (future time point) earlier than the time point at which the prediction process is executed, depending on the embodiment. May be appropriately selected.
  • a learning model such as a decision tree, a neural network, or a support vector man may be used.
  • the “first data” may be data relating to all types of factors that can determine the operation of the target device.
  • the “second data” may be configured by a value that directly designates a command value to the target device (that is, the command value itself), for example, a command value such as a correction value for the reference value of the command value. You may comprise by the value which designates a value indirectly.
  • the threshold value determination unit may employ an estimated boundary value of the numerical range or a value between the first threshold value and the boundary value as the second threshold value.
  • a 2nd threshold value can be determined appropriately so that the prediction control which can fully exhibit the performance of a prediction model can be implemented.
  • the first threshold value may be an upper limit value of the first allowable range
  • the threshold value determination unit adopts a value exceeding the upper limit value as the second threshold value. May be.
  • a 2nd threshold value can be determined appropriately so that the prediction control which can fully exhibit the performance of a prediction model can be implemented.
  • the first threshold value may be a lower limit value of the first allowable range
  • the threshold value determination unit adopts a value smaller than the lower limit value as the second threshold value. May be.
  • a 2nd threshold value can be determined appropriately so that the prediction control which can fully exhibit the performance of a prediction model can be implemented.
  • the threshold value determination unit may determine the second threshold value so as to satisfy a preset safety condition.
  • the “safety condition” may be set as appropriate, may be defined by a threshold value, or may be defined by a simulation or a condition for driving an actual machine.
  • the second data may be constituted by a correction value with respect to a reference value of the command value.
  • the control system which can determine appropriately the command value to an object apparatus can be provided using the correction value obtained from a prediction model.
  • the target device may be a production device that produces a product from a workpiece
  • the first data and the input data are an environment for producing the feature quantity of the workpiece and the product, respectively. May be constituted by at least one of the attribute values.
  • the predictive control of the production apparatus which can fully exhibit the performance of a prediction model can be implemented.
  • the “production device” is not particularly limited as long as it is a device that performs some production processing and can be controlled.
  • a press machine an injection molding machine, an NC lathe, an electric discharge machine, a packaging machine, It may be a transport machine, a transport mechanism in an inspection machine, or the like.
  • the “workpiece” is not particularly limited as long as it can be a work target of the production apparatus.
  • the “workpiece” may be a raw material of a product, a product before processing, a part before assembly, or the like.
  • the “product” is a product obtained by a production apparatus performing a production process on a workpiece, and may include an intermediate product (processed product) in addition to the final product.
  • the “work feature amount” is not particularly limited as long as it can show some feature of the work, and may be appropriately selected according to the embodiment.
  • the feature amount of the workpiece may indicate, for example, hardness, dimensions, material, weight, heat, and the like. Further, the feature amount of the workpiece may directly indicate the feature of the workpiece, or may indirectly indicate the feature of the workpiece. Directly indicating the characteristics of the workpiece means, for example, expressing the hardness (hardness) of the workpiece itself by a numerical value, a class, or the like.
  • indirectly indicating the characteristics of the workpiece means, for example, a secondary index obtained when measuring the hardness (hardness) of the workpiece (for example, a load applied to the workpiece, a torque applied during the measurement) Etc.) are expressed by numerical values, classes, etc.
  • the “attribute value of the environment for producing the product” is not particularly limited as long as it can indicate some attribute relating to the environment in which the production apparatus operates, and may be appropriately selected according to the embodiment.
  • the attribute value of the environment in which the product is produced may indicate, for example, the ambient temperature and humidity of the production apparatus, the degree of deterioration of the apparatus (for example, age, number of processings, etc.), vibration, and the like.
  • one aspect of the present invention may be an information processing method that realizes each of the above configurations, a program, or such It may be a computer-readable storage medium that stores the program.
  • the computer-readable storage medium is a medium that stores information such as programs by electrical, magnetic, optical, mechanical, or chemical action.
  • one aspect of the present invention includes a part of each of the above components (for example, a part for constructing a prediction model, a part for determining a second threshold, a prediction model, and An information processing system that implements the second threshold, etc.), an information processing apparatus, a program, or a computer storing such a program May be a readable storage medium.
  • the computer includes first data relating to a factor for determining an operation of the target device, and a command value to the target device, the factor indicated by the first data.
  • a step of constructing a prediction model so as to output a value corresponding to the second data, and a numerical range that the command value can take is estimated from the distribution of the second data in the acquired plurality of learning data sets.
  • a first allowable range defined by a step and a first threshold value set in advance for the command value to the target device.
  • an output value is obtained from the prediction model, and based on the obtained output value, within a second allowable range defined by the second threshold value determined.
  • an information processing method for executing a step of determining a command value for the target device and a step of controlling an operation of the target device based on the determined command value.
  • the learning device includes first data related to a factor that determines an operation of the target device, and a command value to the target device, the factor indicated by the first data
  • a learning data acquisition unit that acquires a plurality of learning data sets each configured by a combination of second data related to the adapted command value, and the first data is input for each of the acquired learning data sets
  • a learning processing unit that builds a prediction model so as to output a value corresponding to the second data, and a numerical value that the command value can take from the distribution of the second data in the acquired plurality of learning data sets
  • An estimation unit for estimating a range, and a first allowable range defined by a first threshold value set in advance for the command value to the target device is expanded. As provided based on the value range estimated, and a threshold determination unit that determines a second threshold value for the command value for the target device.
  • control device includes an input data acquisition unit that acquires input data related to a factor that determines the operation of the target device, and the acquired input data that is input to the prediction model, An output value is obtained from the prediction model, and a command to the target device is within a second allowable range defined by the second threshold value determined by the learning device according to the configuration based on the obtained output value.
  • a prediction calculation unit that determines a value; and an operation control unit that controls the operation of the target device based on the determined command value.
  • the computer is first data related to a factor that determines the operation of the target device, and a command value to the target device, and is indicated by the first data.
  • a step of constructing a prediction model so as to output a value corresponding to the second data, and a numerical range that the command value can take from the distribution of the second data in the acquired plurality of learning data sets are obtained.
  • a learning program is a computer that includes first data relating to factors that determine the operation of the target device, and a command value to the target device, which is indicated by the first data.
  • a step of constructing a prediction model so as to output a value corresponding to the second data, and a numerical range that the command value can take from the distribution of the second data in the acquired plurality of learning data sets are obtained.
  • a first step that is defined by a first threshold value that is set in advance with respect to the command value to the target device; As expanded containers range, based on the value range estimated, for executing the steps of: determining a second threshold value for the command value for the target device, a program.
  • FIG. 1 schematically illustrates an example of a scene to which the present invention is applied.
  • FIG. 2 schematically illustrates an example of a hardware configuration of the learning device according to the embodiment.
  • FIG. 3 schematically illustrates an example of a hardware configuration of the control device according to the embodiment.
  • FIG. 4 schematically illustrates an example of the production apparatus according to the embodiment.
  • FIG. 5A schematically illustrates an example of a production process in the production apparatus of FIG.
  • FIG. 5B schematically illustrates an example of a production process in the production apparatus of FIG.
  • FIG. 5C schematically illustrates an example of a production process in the production apparatus of FIG.
  • FIG. 5D schematically illustrates an example of a production process in the production apparatus of FIG.
  • FIG. 5A schematically illustrates an example of a production process in the production apparatus of FIG.
  • FIG. 5B schematically illustrates an example of a production process in the production apparatus of FIG.
  • FIG. 5C schematically illustrates an example of a production process in the
  • FIG. 6 schematically illustrates an example of the software configuration of the learning device according to the embodiment.
  • FIG. 7A schematically illustrates an example of a prediction model according to the embodiment.
  • FIG. 7B schematically illustrates the relationship between input and output for the prediction model.
  • FIG. 8 schematically illustrates an example of the software configuration of the control device according to the embodiment.
  • FIG. 9 illustrates an example of a processing procedure of the learning device according to the embodiment.
  • FIG. 10 schematically illustrates an example of the distribution of the second data related to the command value.
  • FIG. 11A schematically illustrates an example of a method for determining the second threshold.
  • FIG. 11B schematically illustrates an example of a method for determining the second threshold.
  • FIG. 12 illustrates an example of a processing procedure of the control device according to the embodiment.
  • this embodiment will be described with reference to the drawings.
  • this embodiment described below is only an illustration of the present invention in all respects. It goes without saying that various improvements and modifications can be made without departing from the scope of the present invention. That is, in implementing the present invention, a specific configuration according to the embodiment may be adopted as appropriate.
  • data appearing in this embodiment is described in a natural language, more specifically, it is specified by a pseudo language, a command, a parameter, a machine language, or the like that can be recognized by a computer.
  • FIG. 1 schematically illustrates an example of a usage scene of the control system 100 according to the present embodiment.
  • a control system 100 illustrated in FIG. 1 includes a learning device 1 and a control device 2 connected via a network, and is configured to control the operation of the production device 3.
  • the type of network between the learning device 1 and the control device 2 may be appropriately selected from, for example, the Internet, a wireless communication network, a mobile communication network, a telephone network, and a dedicated network.
  • the learning device 1 and the control device 2 are separate computers.
  • the configuration of the control system 100 may not be limited to such an example.
  • the learning device 1 and the control device 2 may be configured as an integrated computer.
  • Each of the learning device 1 and the control device 2 may be composed of a plurality of computers.
  • the learning device 1 is a computer configured to construct a prediction model (prediction model 5 described later) for predictive control of the operation of the production device 3.
  • the production apparatus 3 is configured to produce a product from a workpiece, and is an example of the “target apparatus” in the present invention.
  • the “target device” of the present invention is not limited to such a production device 3 and may include all types of devices that can be controlled.
  • the production apparatus 3 is a press machine which processes a workpiece
  • the production apparatus to which the control device 2 can be applied is not limited to such a press, and may be appropriately selected according to the embodiment.
  • the production apparatus 3 may be, for example, an injection molding machine, an NC lathe, an electric discharge machine, a packaging machine, a transport machine, a transport mechanism in an inspection machine, in addition to a press machine.
  • the learning device 1 acquires a plurality of learning data sets (a learning data set 121 described later) in order to construct a prediction model.
  • the plurality of learning data sets are first data (features 1211 and attribute values 1212 described later) relating to factors that determine the operation of the production apparatus 3, and command values to the production apparatus 3, and are based on the first data.
  • Each is constituted by a combination of second data (correction value 1213 to be described later) related to the command value adapted to the indicated factor.
  • the learning device 1 constructs a prediction model so as to output a value corresponding to the second data when the first data is input for each of the plurality of acquired learning data sets.
  • the learning device 1 estimates a numerical range that the command value can take from the distribution of the second data in the acquired plurality of learning data sets. And the learning apparatus 1 is based on the estimated numerical range so that the 1st tolerance
  • the control device 2 is a computer configured to control the operation of the production device 3 using the prediction model constructed by the learning device 1. Specifically, the control device 2 according to the present embodiment acquires input data (a feature amount 71 and an attribute value 72 to be described later) relating to factors that determine the operation of the production device 3. Subsequently, the control device 2 acquires the output value from the prediction model by inputting the acquired input data to the prediction model. Next, the control device 2 determines a command value for the production device 3 within a second allowable range defined by the second threshold value determined by the learning device 1 based on the acquired output value. And the control apparatus 2 controls operation
  • input data a feature amount 71 and an attribute value 72 to be described later
  • the second threshold value set to expand the first allowable range instead of the first allowable range defined by the preset first threshold value.
  • the second allowable range defined by is used as a constraint value for the command value.
  • each learning data set is collected so as to realize operation control suitable for a specific case. Therefore, according to the command value specified based on the second data in each learning data set, the operation of the production apparatus 3 can be controlled safely. Therefore, based on the numerical range estimated from the distribution of the second data in the learning data set, the second threshold value defining the second allowable range is set so as to ensure the safety of the operation of the production apparatus 3. Can be determined. Therefore, according to the control system 100 according to the present embodiment, it is possible to perform predictive control that can sufficiently exhibit the performance of the predictive model while ensuring the safety of the operation of the production apparatus 3.
  • FIG. 2 schematically illustrates an example of a hardware configuration of the learning device 1 according to the present embodiment.
  • the learning device 1 is a computer in which a control unit 11, a storage unit 12, a communication interface 13, an input device 14, an output device 15, and a drive 16 are electrically connected.
  • the communication interface is described as “communication I / F”.
  • the control unit 11 includes a CPU (Central Processing Unit), a RAM (Random Access Memory), a ROM (Read Only Memory), etc., which are hardware processors, and is configured to execute information processing based on programs and various data.
  • the storage unit 12 is an example of a memory, and includes, for example, a hard disk drive or a solid state drive. In the present embodiment, the storage unit 12 stores various information such as a learning program 81 executed by the control unit 11 (CPU), a plurality of learning data sets 121, and learning result data 125.
  • the learning program 81 is a program for causing the learning apparatus 1 to execute later-described information processing (FIG. 9) of machine learning for constructing a prediction model and generating learning result data 125 as a result of the machine learning.
  • the learning program 81 includes a series of instructions for the information processing.
  • the plurality of learning data sets 121 are data used for machine learning for constructing a prediction model that has acquired the ability to predict a command value adapted to product production by the production apparatus 3. Details will be described later.
  • the communication interface 13 is, for example, a wired LAN (Local Area Network) module, a wireless LAN module, or the like, and is an interface for performing wired or wireless communication via a network.
  • the learning device 1 can perform data communication via the network with another information processing device (for example, the control device 2) by using the communication interface 13.
  • the learning device 1 can distribute the generated learning result data 125 to an external device by using the communication interface 13.
  • the input device 14 is a device for inputting, for example, a mouse and a keyboard.
  • the output device 15 is a device for outputting, for example, a display or a speaker. The operator can operate the learning device 1 by using the input device 14 and the output device 15.
  • the drive 16 is, for example, a CD drive, a DVD drive, or the like, and is a drive device for reading a program stored in the storage medium 91.
  • the type of the drive 16 may be appropriately selected according to the type of the storage medium 91.
  • At least one of the learning program 81 and the learning data set 121 may be stored in the storage medium 91.
  • the storage medium 91 stores information such as a program by electrical, magnetic, optical, mechanical, or chemical action so that a computer or other device, machine, or the like can read the recorded program or the like. It is a medium to accumulate.
  • the learning device 1 may acquire at least one of the learning program 81 and the learning data set 121 from the storage medium 91.
  • a disk-type storage medium such as a CD or a DVD is illustrated.
  • the type of the storage medium 91 is not limited to the disk type and may be other than the disk type.
  • Examples of the storage medium other than the disk type include a semiconductor memory such as a flash memory.
  • the control unit 11 may include a plurality of hardware processors.
  • the hardware processor may be configured by a microprocessor, an FPGA (field-programmable gate array), or the like.
  • the storage unit 12 may be configured by a RAM and a ROM included in the control unit 11. At least one of the communication interface 13, the input device 14, the output device 15, and the drive 16 may be omitted.
  • the learning device 1 may be composed of a plurality of information processing devices. In this case, the hardware configurations of the computers may or may not match.
  • the learning device 1 may be a general-purpose server device, a general-purpose PC (Personal Computer), or the like, in addition to an information processing device designed exclusively for the provided service.
  • FIG. 3 schematically illustrates an example of a hardware configuration of the control device 2 according to the present embodiment.
  • the control device 2 includes a control unit 21, a storage unit 22, a communication interface 23, an external interface 24, an input device 25, an output device 26, and a drive 27 that are electrically connected.
  • Computer In FIG. 3, the communication interface and the external interface are described as “communication I / F” and “external I / F”, respectively.
  • control unit 21 includes a CPU, RAM, ROM, and the like, which are hardware processors, and is configured to execute information processing based on programs and various data.
  • the storage unit 22 is configured by, for example, a hard disk drive, a solid state drive, or the like.
  • the storage unit 22 stores various information such as a control program 82 executed by the control unit 21 (CPU) and learning result data 125.
  • the control program 82 is a program for causing the control device 2 to execute information processing (FIG. 12) described later for controlling the operation of the production device 3, and includes a series of instructions for the information processing.
  • the learning result data 125 is data for setting a learned prediction model. Details will be described later.
  • the communication interface 23 is, for example, a wired LAN module, a wireless LAN module, or the like, and is an interface for performing wired or wireless communication via a network.
  • the control device 2 can perform data communication via the network with another information processing device (for example, the learning device 1).
  • the external interface 24 is, for example, a USB (Universal Serial Bus) port, a dedicated port, or the like, and is an interface for connecting to an external device.
  • the type and number of external interfaces 24 may be appropriately selected according to the type and number of external devices to be connected.
  • the control device 2 is connected to the production device 3 via the external interface 24. Thereby, the control apparatus 2 can control the operation of the production apparatus 3 by transmitting a command value to the production apparatus 3.
  • the input device 25 is a device for inputting, for example, a mouse and a keyboard.
  • the output device 26 is a device for outputting, for example, a display or a speaker. An operator can operate the control device 2 via the input device 25 and the output device 26.
  • the drive 27 is, for example, a CD drive, a DVD drive, or the like, and is a drive device for reading a program stored in the storage medium 92.
  • the type of the drive 27 may be appropriately selected according to the type of the storage medium 92.
  • At least one of the control program 82 and the learning result data 125 may be stored in the storage medium 92.
  • the storage medium 92 stores information such as a program by an electrical, magnetic, optical, mechanical, or chemical action so that information such as a program recorded by a computer or other device or machine can be read. It is a medium to do.
  • the control device 2 may acquire at least one of the control program 82 and the learning result data 125 from the storage medium 92.
  • FIG. 3 illustrates a disk-type storage medium such as a CD and a DVD as an example of the storage medium 92 as in FIG.
  • the type of the storage medium 92 is not limited to the disk type and may be other than the disk type.
  • Examples of the storage medium other than the disk type include a semiconductor memory such as a flash memory.
  • control unit 21 may include a plurality of hardware processors.
  • the hardware processor may be configured by a microprocessor, FPGA, DSP, or the like.
  • the storage unit 22 may be configured by a RAM and a ROM included in the control unit 21. At least one of the communication interface 23, the external interface 24, the input device 25, the output device 26, and the drive 27 may be omitted.
  • the control device 2 may be composed of a plurality of computers. In this case, the hardware configurations of the computers may or may not match.
  • the control device 2 may be a general-purpose controller, a general-purpose server device, a general-purpose desktop PC, a notebook PC, a tablet PC, or the like, in addition to an information processing device designed exclusively for the provided service.
  • FIG. 4 schematically illustrates an example of a hardware configuration of the production apparatus 3 according to the present embodiment.
  • the production apparatus 3 includes a servo driver 31, an upper mold 32, and a lower mold 33. While the lower mold 33 is fixed, the upper mold 32 is configured to be movable in the vertical direction by a servo motor (not shown). As a result, the upper mold 32 can press the workpiece against the lower mold 33 to mold the workpiece, or to move away from the lower mold 33.
  • the servo driver 31 is configured to drive the servo motor of the upper mold 32 based on the command value from the control device 2.
  • the production apparatus 3 is arranged on a production line, for example.
  • the upper mold 32 is arranged at a standby position away from the lower mold 33 and waits until the workpiece 40 is conveyed to the lower mold 33.
  • the workpiece 40 is, for example, a metal plate material.
  • the workpiece 40 is not limited to such an example, and may be appropriately selected according to the type of the production apparatus 3.
  • the workpiece 40 may be, for example, a raw material of a product, a product before processing, a part before assembly, or the like.
  • the production apparatus 3 drives the servo motor of the upper mold 32 by the servo driver 31, as shown in FIG. Place at the molding start position.
  • the molding start position is, for example, a position where the tip of the upper mold 32 comes into contact with the workpiece 40 or immediately before it.
  • the production apparatus 3 further drives the servo motor of the upper mold 32 by the servo driver 31, moves the upper mold 32 to the target position (bottom dead center), and moves the upper mold 32. Then, the workpiece 40 is formed by the lower mold 33. Thereby, the production apparatus 3 can produce the product 41 from the workpiece 40.
  • the product 41 is not particularly limited as long as it is a product obtained by the production apparatus 3 performing a production process on the workpiece 40, and may be a final product or an intermediate product (during processing). May be).
  • the production apparatus 3 drives the servo motor of the upper mold 32 by the servo driver 31, as shown in FIG. 5D, and moves the upper mold 32 to the standby position. And the product 41 obtained by shape
  • the control device 2 predicts an appropriate command value to the production device 3 by using the prediction model so that no defect occurs in the production process. Thereby, the control apparatus 2 adjusts operation
  • FIG. 6 schematically illustrates an example of the software configuration of the learning device 1 according to the present embodiment.
  • the control unit 11 of the learning device 1 expands the learning program 81 stored in the storage unit 12 in the RAM. Then, the control unit 11 interprets and executes the learning program 81 expanded in the RAM, and controls each component based on a series of instructions included in the learning program 81. Accordingly, as illustrated in FIG. 6, the learning device 1 according to the present embodiment operates as a computer including the learning data acquisition unit 111, the learning processing unit 112, the estimation unit 113, and the threshold value determination unit 114 as software modules. That is, in the present embodiment, each software module is realized by the control unit 11 (CPU).
  • the learning data acquisition unit 111 acquires a plurality of learning data sets 121 used for machine learning of the prediction model 5.
  • Each learning data set 121 includes first data relating to factors that determine the operation of the production apparatus 3 and command values to the production apparatus 3, which are command values adapted to the factors indicated by the first data. It consists of a combination of two data.
  • the first data is composed of the feature value 1211 of the work 40 and the attribute value 1212 of the environment in which the product 41 is produced.
  • the second data is a correction value 1213 with respect to the reference value of the command value, and is composed of a correction value 1213 determined so that a command value adapted to the situation indicated by the feature value 1211 and the attribute value 1212 can be obtained.
  • the first data corresponds to training data (input data), and the second data corresponds to teacher data (correct answer data).
  • the learning processing unit 112 constructs the learned prediction model 5 by performing machine learning using the plurality of acquired learning data sets 121. That is, when the learning processing unit 112 inputs the first data (feature value 1211 and attribute value 1212) for each acquired learning data set 121, the second data (correction) associated with the input first data is input. The prediction model 5 is constructed so as to output a value corresponding to the value 1213). Then, the learning processing unit 112 stores information on the constructed learned prediction model 5 in the storage unit 12 as learning result data 125.
  • the estimation unit 113 estimates a possible numerical range of the command value from the distribution 61 of the second data (correction value 1213) in the acquired plurality of learning data sets 121. Then, the threshold value determination unit 114 performs the production based on the estimated numerical range so as to expand the first allowable range defined by the first threshold value 60 set in advance with respect to the command value to the production apparatus 3. A second threshold 62 for the command value to the device 3 is determined.
  • FIG. 7A schematically illustrates an example of the configuration of the prediction model 5 according to the present embodiment.
  • FIG. 7B schematically illustrates the relationship between the input and output for the prediction model 5.
  • the prediction model 5 is configured by a decision tree (specifically, a regression tree).
  • the prediction model 5 includes a root node R, leaf nodes L1 to L5, and intermediate nodes N1 to N3 arranged between the root node R and the leaf nodes L1 to L5.
  • a link is provided between each node.
  • the root node R and the intermediate nodes (N1, N2) between the intermediate node N1 and each leaf node (L1, L2), between the intermediate node N2, the leaf node L3, and the intermediate node N3.
  • a link is provided between the intermediate node N3 and each leaf node (L4, L5).
  • the depth of the decision tree is 4, the number of intermediate nodes is 3, and the number of leaf nodes is 5.
  • the depth of the decision tree, the number of intermediate nodes, and the number of leaf nodes do not have to be limited to such an example, and may be appropriately determined according to the embodiment.
  • no link is provided from the root node R to each of the leaf nodes L1 to L5.
  • the configuration of the decision tree may not be limited to such an example, and there may be leaf nodes connected to the link from the root node.
  • the calculation process of the prediction model 5 is a search process that follows links from the root node R of the decision tree toward the leaf nodes L1 to L5. That is, a branch condition is associated with the route from the root node R to the leaf nodes L1 to L5 (in the example of FIG. 7A, the root node R and the intermediate nodes N1 to N3).
  • the root node R has a branch condition “x0 ⁇ 2500”
  • the intermediate node N1 has a branch condition “x1 ⁇ 20”
  • the intermediate node N2 has a branch condition “x1 ⁇ 35”.
  • the branch condition “x0 ⁇ 3500” is associated with the intermediate node N3.
  • the final results (classes C1 to C5) of the calculation process of the prediction model 5 are associated with the leaf nodes L1 to L5.
  • each leaf node L1 to L5 (class C1 to C5) is associated with a correction value corresponding to the input feature value and attribute value. That is, when the learning processing unit 112 inputs the feature value 1211 and the attribute value 1212, the prediction model 5 is reached so as to reach the leaf node of the class corresponding to the correction value 1213 associated with the input feature value 1211 and attribute value 1212. Build a (decision tree).
  • the learning processing unit 112 stores the configuration of the constructed learned prediction model 5, information indicating each branch condition, and the like in the storage unit 12 as learning result data 125.
  • FIG. 8 schematically illustrates an example of the software configuration of the control device 2 according to the present embodiment.
  • the control unit 21 of the control device 2 expands the control program 82 stored in the storage unit 22 in the RAM. Then, the control unit 21 interprets and executes the control program 82 expanded in the RAM, and controls each component based on a series of instructions included in the control program 82. Accordingly, as illustrated in FIG. 8, the control device 2 according to the present embodiment operates as a computer including the input data acquisition unit 211, the prediction calculation unit 212, and the operation control unit 213 as software modules. That is, in this embodiment, each software module is realized by the control unit 21 (CPU).
  • the input data acquisition unit 211 acquires input data relating to factors that determine the operation of the production apparatus 3.
  • the prediction model 5 is constructed so as to predict a command value adapted to the production of the product 41 with respect to the input of the feature value of the workpiece 40 and the attribute value of the environment in which the product 41 is produced. Therefore, the input data acquisition unit 211 acquires the feature value 71 of the workpiece 40 and the attribute value 72 of the environment in which the product 41 is produced as input data.
  • the prediction calculation unit 212 holds the learning result data 125 generated by the learning device 1. Thereby, the prediction calculation unit 212 is a command value that is a command value from the workpiece 40 to the production apparatus 3 that produces the product 41, and is configured to predict a command value that is adapted to the production of the product 41 by the production apparatus 3. Model 5 is provided. The prediction calculation unit 212 refers to the learning result data 125 and sets the prediction model 5 used for prediction control.
  • the prediction calculation unit 212 inputs the acquired input data (feature value 71 and attribute value 72) to the prediction model 5, and executes the calculation process of the prediction model 5. Thereby, the prediction calculation unit 212 acquires from the prediction model 5 an output value corresponding to a result of predicting a command value adapted to the production of the product 41 by the production apparatus 3. The prediction calculation unit 212 determines a command value for the production apparatus 3 within the second allowable range defined by the second threshold value 62 determined by the learning apparatus 1 based on the acquired output value.
  • the prediction model 5 is configured by a decision tree that outputs a correction value 73 for the reference value 70 of the command value as an output value corresponding to the result of prediction of the command value adapted to the production of the product 41. . Therefore, the prediction calculation unit 212 executes a decision tree search process as the calculation process of the prediction model 5. The prediction calculation unit 212 can acquire an output value corresponding to the correction value 73 from the prediction model 5 by completing the calculation process of the prediction model 5.
  • FIG. 7A a search process of the decision tree (prediction model 5) illustrated in FIG. 7A will be described.
  • the prediction calculation unit 212 starts the search process from the root node R of the prediction model 5 and repeats the determination of whether or not the input data satisfies the branch condition until reaching any one of the leaf nodes L1 to L5. , The search proceeds to deeper nodes.
  • the input x0 corresponds to the feature amount 71
  • the input x1 corresponds to the attribute value 72.
  • FIG. 7B illustrates the relationship between each input (x0, x1) and the classes C1 to C5 associated with the reaching leaf nodes L1 to L5.
  • the prediction calculation unit 212 determines whether or not the input x0 satisfies the branch condition set in the root node R as the calculation processing (search processing) in the first layer of the prediction model 5.
  • search processing search processing
  • the prediction calculation unit 212 inputs the branch condition set for the root node R to the input x0. Is satisfied and the search proceeds to the intermediate node N1 of the next hierarchy.
  • the prediction calculation unit 212 determines whether or not the input x1 satisfies the branch condition set in the intermediate node N1 as the calculation processing of the second hierarchy of the prediction model 5.
  • the prediction calculation unit 212 since the branch condition set in the intermediate node N1 is “x1 ⁇ 20” and the input x1 is 30, the prediction calculation unit 212 inputs the branch condition set in the intermediate node N1 as input x1. Is not satisfied, and the process proceeds to the leaf node L2 of the next hierarchy. Thereby, since the search process of the decision tree reaches the leaf node L2, the calculation process of the prediction model 5 is completed.
  • the prediction calculation unit 212 can acquire the correction value 73 associated with the class C2 of the leaf node L2 as the final result of the calculation process of the prediction model 5.
  • the method for acquiring the correction value 73 associated with each of the classes C1 to C5 may be appropriately determined according to the embodiment.
  • correction values may be directly associated with the classes C1 to C5.
  • the control device 2 may hold reference information such as a table format indicating the correspondence between the classes C1 to C5 and the correction values in the storage unit 22. This reference information may be generated in the learning process of the prediction model 5 and may be included in the learning result data 125.
  • the prediction calculation unit 212 collates the class of the reached leaf node with the reference information, thereby obtaining the reference value of the command value as the final result of the calculation process of the prediction model 5
  • a correction value 73 for 70 can be acquired.
  • the prediction calculation unit 212 determines the command value 75 within the second allowable range based on the value obtained by correcting the reference value 70 with the acquired correction value 73.
  • the prediction calculation unit 212 determines the obtained value as the command value 75.
  • the prediction calculation unit 212 corrects the obtained value to change the second value.
  • the command value 75 is determined within the allowable range.
  • the operation control unit 213 controls the operation of the production apparatus 3 based on the determined command value 75.
  • each software module of the learning device 1 and the control device 2 is realized by a general-purpose CPU.
  • some or all of the above software modules may be implemented by one or more dedicated processors.
  • software modules may be omitted, replaced, and added as appropriate according to the embodiment.
  • FIG. 9 is a flowchart illustrating an example of a processing procedure of the learning device 1.
  • the processing procedure described below is an example of the “learning method” of the present invention. However, the processing procedure described below is merely an example, and each processing may be changed as much as possible. Further, in the processing procedure described below, steps can be omitted, replaced, and added as appropriate according to the embodiment.
  • Step S101 In step S ⁇ b> 101, the control unit 11 operates as the learning data acquisition unit 111 and acquires a plurality of learning data sets 121 used for machine learning of the prediction model 5.
  • Each learning data set 121 includes first data relating to factors that determine the operation of the production apparatus 3 and command values to the production apparatus 3, which are command values adapted to the factors indicated by the first data. It consists of a combination of two data.
  • the first data includes the feature value 1211 of the workpiece 40 and the attribute value 1212 of the environment in which the product 41 is produced.
  • the second data is a correction value 1213 with respect to the reference value of the command value, and is composed of a correction value 1213 determined so that a command value adapted to the situation indicated by the feature value 1211 and the attribute value 1212 can be obtained.
  • the feature amount 1211 of the workpiece 40 may not be particularly limited as long as it can show some feature of the workpiece 40, and may be appropriately selected according to the embodiment.
  • the attribute value 1212 of the environment in which the product 41 is produced may not be particularly limited as long as it can indicate some attribute relating to the environment in which the production apparatus 3 operates, and may be appropriately selected according to the embodiment. .
  • the production apparatus 3 is a press machine. As described above, in the production apparatus 3, if the press time is insufficient or the servo motor is not driven until the upper die 32 reaches the bottom dead center, the quality of the product 41 obtained is deteriorated. End up. For this reason, it is preferable that the feature value 1211 of the workpiece 40 and the attribute value 1212 of the environment for producing the product 41 each relate to a press forming process in the production apparatus 3.
  • the feature quantity 1211 of the workpiece 40 for example, a value indicating hardness, dimensions, material, weight, heat, or the like may be selected.
  • the attribute value 1212 of the environment in which the product 41 is produced for example, a value indicating the ambient temperature and humidity of the production apparatus 3, the degree of deterioration of the apparatus (for example, age, the number of times of processing, etc.), vibration, etc. It's okay.
  • the feature amount 1211 of the workpiece 40 may directly indicate the feature of the workpiece 40 or may indirectly indicate the feature of the workpiece 40.
  • Directly indicating the characteristics of the workpiece 40 is, for example, expressing the hardness (hardness) of the workpiece 40 itself as a numerical value, a class, or the like.
  • indirectly indicating the characteristics of the workpiece 40 is, for example, a secondary index obtained when measuring the hardness (hardness) of the workpiece 40 (for example, acting on the workpiece, the load applied to the workpiece, and the measurement). Torque, etc.) is expressed by numerical value, class, etc. The same applies to the attribute value 1212.
  • Such a learning data set 121 for each case may be appropriately generated according to the embodiment.
  • the production apparatus 3 is operated, and the feature value 1211 of the workpiece 40 and the attribute value 1212 of the environment for producing the product 41 are acquired under various conditions.
  • a known sensor may be used to acquire the feature amount 1211 and the attribute value 1212.
  • a hardness meter may be used.
  • a temperature sensor may be used. Then, the obtained feature value 1211 and attribute value 1212 are combined with a correction value 1213 for obtaining an appropriate command value under this condition. Thereby, each learning data set 121 can be generated.
  • the generation of the learning data set 121 may be performed by the learning device 1.
  • the control unit 11 may generate each learning data set 121 in accordance with the operation of the input device 14 by the operator. Further, the control unit 11 may automatically generate the learning data set 121 for each case by the processing of the learning program 81. By executing this generation process, in step S101, the control unit 11 can acquire a plurality of learning data sets 121.
  • the generation of the learning data set 121 may be performed by an information processing apparatus other than the learning apparatus 1.
  • the learning data set 121 for each case may be manually generated by an operator or may be automatically generated by processing of a program.
  • the control unit 11 can acquire a plurality of learning data sets 121 generated by other information processing apparatuses via the network, the storage medium 91, and the like.
  • step S101 the number of learning data sets 121 acquired in step S101 may be determined as appropriate according to the embodiment, and may be determined as appropriate to the extent that machine learning of a decision tree can be performed, for example.
  • the control unit 11 advances the processing to the next step S102.
  • Step S102 In step S ⁇ b> 102, the control unit 11 operates as the learning processing unit 112 and constructs the learned prediction model 5 by performing machine learning using the plurality of acquired learning data sets 121.
  • the control unit 11 when the control unit 11 inputs the feature value 1211 and the attribute value 1212 for the acquired learning data set 121, the control unit 11 sets the correction value 1213 associated with the input feature value 1211 and attribute value 1212.
  • the prediction model 5 is constructed so as to output a corresponding value. More specifically, the control unit 11 starts from the root node based on the feature value 1211 and the attribute value 1212 and determines a decision tree that can be searched to reach the leaf node of the class corresponding to the associated correction value 1213. To construct. For this decision tree learning method, CLS (ConceptConLearning System), ID3 (Iterative Dichotomisertom3), C4.5, or the like may be used. Thereby, the control unit 11 can construct the learned prediction model 5. When the learned prediction model 5 is constructed, the control unit 11 advances the processing to the next step S103.
  • Step S ⁇ b> 103 the control unit 11 operates as the estimation unit 113 and estimates a numerical value range that the command value can take from the distribution 61 of the second data in the acquired plurality of learning data sets 121.
  • step S104 the control unit 11 operates as the threshold value determination unit 114, and expands the first allowable range defined by the first threshold value 60 set in advance with respect to the command value to the production apparatus 3. Based on the estimated numerical range, the second threshold value 62 for the command value to the production apparatus 3 is determined.
  • the expression range of the estimated command value, the first threshold value 60, and the second threshold value 62 may not be particularly limited, and may be appropriately determined according to the embodiment. It's okay.
  • the second data is constituted by a correction value 1213. Therefore, in step S103, the control unit 11 may indirectly estimate the numerical range that the command value can take by estimating the numerical range that the correction value 1213 can take. Accordingly, the first threshold value 60 and the second threshold value 62 may be set for the correction value to indirectly define the allowable range of the command value. Further, the control unit 11 may directly estimate the numerical range that the command value can take, based on a value obtained by correcting the reference value 70 with the correction value 1213.
  • the first threshold 60 and the second threshold 62 may be set for the command value, so that the allowable range of the command value may be set directly. Both cases can be handled in the same way.
  • a numerical value range that the command value can take is directly estimated, and the first threshold value 60 and the second threshold value 62 are set directly with respect to the command value. Assumes something.
  • the control unit 11 can grasp the distribution of the command value specified by the correction value 1213 (second data) by referring to the correction value 1213 (second data) in the learning data set 121 of each case. it can. At this time, the control unit 11 may approximate the distribution of the command value using a statistical method such as a normal distribution, a gamma distribution, and an exponential distribution.
  • FIG. 10 shows an example of a scene in which the distribution of command values is approximated by a normal distribution.
  • a known statistical process may be used as a method of approximating the distribution of command values by a normal distribution.
  • the control unit 11 can calculate the minimum value and the maximum value of the command value based on the approximated normal distribution.
  • the control unit 11 may estimate a numerical range from the minimum value to the maximum value as a numerical range that the command value can take.
  • Each of the minimum value and the maximum value of the command value in the normal distribution is an example of a boundary value in a numerical range that can be taken by the command value.
  • the method of estimating the numerical range that the command value can take may not be limited to such a statistical method.
  • the control unit 11 uses the distribution obtained by referring to the correction value 1213 (second data) in the learning data set 121 of each case as it is as a numerical range that the command value can take. May be.
  • the minimum value and the maximum value of the command value specified by the correction value 1213 (second data) of the learning data set 121 are respectively boundary values of numerical values that can be taken by the command value.
  • the control unit 11 may estimate the numerical range from the minimum value to the maximum value as a numerical range that the command value can take.
  • (C) Second Threshold Determination Method a method for determining the second threshold 62 based on the estimated numerical range in step S104 will be described.
  • the method for deriving the second threshold value 62 from the numerical range estimated in step S103 may be set as appropriate according to the embodiment.
  • the control unit 11 can determine the second threshold 62 by using the boundary value of the numerical value range.
  • the control unit 11 may adopt a boundary value in the estimated numerical range or a value between the first threshold value 60 and the boundary value as the second threshold value 62.
  • the allowable range of the command value can be defined by specifying at least one of the lower limit value and the upper limit value.
  • the first threshold 60 may be a lower limit value of the first allowable range, or may be an upper limit value of the first allowable range.
  • each of the lower limit value and the upper limit value of the first allowable range may be handled as the first threshold value 60.
  • FIG. 11A and FIG. 11B below it is assumed that the lower limit value and the upper limit value of the first allowable range are each handled as the first threshold value 60 for convenience of explanation.
  • FIG. 11A schematically illustrates an example of a scene in which the estimated boundary value of the numerical range is adopted as the second threshold value 62.
  • FIG. 11B schematically illustrates an example of a scene in which a value between the estimated boundary value of the numerical range and the first threshold value 60 is adopted as the second threshold value 62.
  • the horizontal axis of the graph corresponds to the command value (input) to the production apparatus 3
  • the vertical axis of the graph corresponds to the torque (output) of the servo motor.
  • the first threshold value 60 is given in advance before the second threshold value 62 is determined.
  • the first threshold 60 may be determined in advance by a user who uses the production apparatus 3, or may be determined in advance in the production apparatus 3 or the control apparatus 2.
  • the first threshold value 60 is used as a restriction condition for the command value of the production apparatus 3 in a scene where the user manually operates the production apparatus 3 or the control apparatus 2 instead of the scene where the production apparatus 3 is controlled by the prediction model 5. May be.
  • the control unit 11 may acquire the first threshold 60 by making an inquiry to the control device 2 or the production device 3 via a network or the like.
  • the learning apparatus 1 may hold
  • step S104 when the maximum value of the numerical range estimated in step S103 exceeds the upper limit value (first threshold 60) of the first allowable range, in step S104, the control unit 11 determines the maximum value. Is adopted as the second threshold 62. Accordingly, the control unit 11 can employ a value that exceeds the upper limit value of the first allowable range as the second threshold value 62.
  • the second threshold 62 is handled as the upper limit value of the second allowable range.
  • step S104 when the minimum value of the numerical range estimated in step S103 is less than the lower limit value (first threshold 60) of the first allowable range, in step S104, the control unit 11 sets the minimum value to the second value.
  • the threshold 62 is adopted. Thereby, the control unit 11 can employ a value smaller than the lower limit value of the first allowable range as the second threshold value 62.
  • the second threshold 62 is handled as the lower limit value of the second allowable range.
  • step S104 when the maximum value of the numerical range estimated in step S103 exceeds the upper limit value (first threshold 60) of the first allowable range, in step S104, the control unit 11 A value between the maximum value and the upper limit value is adopted as the second threshold value 62.
  • the value employed as the second threshold 62 may be appropriately determined according to the embodiment.
  • the control unit 11 may adopt an average value of the estimated maximum value of the numerical range and the upper limit value of the first allowable range as the second threshold value 62. Accordingly, the control unit 11 can employ a value that exceeds the upper limit value of the first allowable range as the second threshold value 62.
  • the second threshold 62 is handled as the upper limit value of the second allowable range.
  • step S104 the control unit 11 determines that the minimum value and the lower limit A value between the values is adopted as the second threshold value 62.
  • the value adopted as the second threshold value 62 may be appropriately determined according to the embodiment.
  • the control unit 11 may adopt an average value of the estimated minimum value of the numerical range and the lower limit value of the first allowable range as the second threshold value 62. Thereby, the control unit 11 can employ a value smaller than the lower limit value of the first allowable range as the second threshold value 62.
  • the second threshold 62 is handled as the lower limit value of the second allowable range.
  • the control unit 11 uses the numerical values estimated in step S103 so as to widen the first allowable range defined by the preset first threshold 60 by any one of the methods described above. Based on the range, the second threshold 62 for the command value can be determined.
  • the control unit 11 may determine the second threshold 62 so as to satisfy a preset safety condition.
  • the safety condition may be appropriately defined according to the embodiment so that the operation of the production apparatus 3 can be safely controlled.
  • the control unit 11 may acquire information indicating the safety condition by making an inquiry to the control device 2 or the production device 3 via a network or the like.
  • the learning apparatus 1 may hold
  • the safety condition may be defined by a threshold value for safety control designated in advance by the user, the manufacturer of the production apparatus 3, or the like.
  • the control unit 11 determines that the value determined in step S104 (the upper limit value of the second allowable range) is equal to or less than the threshold value for safety control. It may be determined whether or not.
  • the control unit 11 adopts the determined value as the second threshold value 62 (that is, the upper limit value of the second allowable range). Also good.
  • the control unit 11 may correct the value so as to be equal to or less than the threshold value for safety control, and adopt the corrected value as the second threshold value 62.
  • the control unit 11 determines that the value determined in step S104 (the lower limit value of the second allowable range) is for safety control. You may determine whether it is more than a threshold value. When the value determined in step S104 is equal to or greater than the threshold value for safety control, the control unit 11 adopts the determined value as the second threshold value 62 (that is, the lower limit value of the second allowable range). Also good. On the other hand, when this is not the case, the control unit 11 may correct the value so as to be equal to or higher than the threshold value for safety control, and adopt the corrected value as the second threshold value 62.
  • the safety condition may be defined by simulating the operation of the production apparatus 3 or actually driving the production apparatus 3.
  • the control unit 11 can safely operate the production apparatus 3 when the value determined as the second threshold value 62 in step S104 is adopted as the command value based on the result of simulation or actual driving. It may be determined whether or not it is possible. And when it determines with the production apparatus 3 being able to operate
  • the control unit 11 corrects the value so that the production apparatus 3 can be operated safely, and sets the corrected value to the second threshold value. You may employ
  • the control unit 11 advances the processing to the next step S105.
  • the control unit 11 determines the upper limit value of the second allowable range (second The process of determining the threshold value 62) may be omitted.
  • the control unit 11 performs a process of determining the second threshold value 62 by any one of the above methods. It may be omitted.
  • the method of determining the upper limit value and the lower limit value of the second allowable range may be different from each other.
  • the control unit 11 adopts the method shown in FIG. 11A as the method for determining the upper limit value of the second allowable range, and uses the method shown in FIG. 11B as the method of determining the lower limit value of the second allowable range. It may be adopted.
  • 11A and 11B schematically illustrate an example of the relationship between the command value (input) to the production apparatus 3 and the torque (output) of the servo motor that drives the upper mold 32.
  • the first threshold value 60 and the second threshold value 62 are set for the command value.
  • the format of the first threshold value 60 and the second threshold value 62 may not be limited to such an example.
  • each of the first threshold value 60 and the second threshold value 62 is set for the output of the target device (in this embodiment, the torque of the servo motor), thereby indirectly specifying the allowable range of the command value. May be.
  • Step S105 the control unit 11 operates as the learning processing unit 112, and stores the information indicating the configuration of the decision tree (learned prediction model 5) constructed by machine learning and each branch condition as learning result data 125. To store.
  • the control unit 11 operates as the threshold value determination unit 114 and stores the second threshold value 62 determined in step S104 in the storage unit 12. Thereby, the control part 11 complete
  • the control unit 11 may transfer the generated learning result data 125 and the second threshold value 62 to the control device 2 after the process of step S105 is completed. Further, the control unit 11 may periodically update the learning result data 125 and the second threshold value 62 by periodically executing the learning process of steps S101 to S105. And the control part 11 transfers the learning result data 125 and the 2nd threshold value 62 which were produced
  • the control unit 11 may store the generated learning result data 125 and the second threshold 62 in a data server such as NAS (Network Attached Storage). In this case, the control device 2 may acquire the learning result data 125 and the second threshold value 62 from this data server. Further, the learning result data 125 and the second threshold value 62 generated by the learning device 1 may be incorporated in the control device 2 in advance.
  • NAS Network Attached Storage
  • FIG. 12 is a flowchart illustrating an example of a processing procedure of the control device 2. Note that the processing procedure described below is merely an example, and each processing may be changed as much as possible. Further, in the processing procedure described below, steps can be omitted, replaced, and added as appropriate according to the embodiment.
  • Step S201 the control unit 21 operates as the input data acquisition unit 211, and acquires input data regarding factors in the operation phase.
  • the prediction model 5 is constructed by machine learning using the learning data set 121 including the feature value 1211 of the workpiece 40 and the attribute value 1212 of the environment in which the product 41 is produced as the first data.
  • the control unit 21 acquires the feature value 71 of the workpiece 40 and the attribute value 72 of the environment in which the product 41 is produced.
  • the feature amount 71 and the attribute value 72 may be the same type as the feature amount 1211 and the attribute value 1212. Further, the method for acquiring the feature amount 71 and the attribute value 72 may be appropriately selected according to the embodiment. For example, various sensors configured to be able to measure the feature amount 71 (for example, hardness) of the workpiece 40 and the environmental attribute value 72 (for example, temperature) may be arranged in the production apparatus 3. As the various sensors, known sensors may be appropriately used according to the types of the feature amount 71 and the attribute value 72 to be measured. In this case, the control unit 21 can acquire each of the feature value 71 and the attribute value 72 from various sensors arranged in the production apparatus 3. When the feature amount 71 and the attribute value 72 are acquired, the control unit 21 proceeds to the next step S202.
  • various sensors configured to be able to measure the feature amount 71 (for example, hardness) of the workpiece 40 and the environmental attribute value 72 (for example, temperature) may be arranged in the production apparatus 3.
  • known sensors may be appropriately used according to the types of the
  • Step S202 the control unit 21 operates as the prediction calculation unit 212, inputs the acquired input data (feature value 71 and attribute value 72) to the prediction model 5, and executes the calculation process of the prediction model 5. Thereby, the control unit 21 acquires an output value corresponding to a result of predicting a command value adapted to the production of the product 41 by the production apparatus 3 from the prediction model 5.
  • the prediction model 5 is configured by a decision tree, and information indicating the configuration of the prediction model 5 and the branch condition of each path is included in the learning result data 125. Therefore, the control unit 21 sets the prediction model 5 by referring to the learning result data 125. By this setting process, the control unit 21 is in a state where the search process for the decision tree (prediction model 5) can be started.
  • control unit 21 executes a search process that follows the link from the root node of the decision tree (prediction model 5) toward the leaf node. Specifically, when the search process has never been executed, the control unit 21 uses the branch condition set in the root node as input data (feature value 71 and attribute value 72) as the search process of the decision tree. It is determined whether or not it is satisfied. Based on the determination result, the control unit 21 advances the search to the corresponding node in the second stage (in the example of FIG. 7A, the intermediate node N1 or the intermediate node N2).
  • the search process is executed n times (n is a natural number of 1 or more), the search is progressing to the intermediate node at the (n + 1) th stage.
  • the control unit 21 determines whether or not the input data satisfies the branch condition set in the corresponding intermediate node in the (n + 1) th stage. Based on the determination result, the control unit 21 advances the search to the corresponding node in the (n + 2) th stage.
  • the control unit 21 can acquire the correction value 73 associated with the leaf node to which the search process has reached as the output value from the prediction model. When the output value is acquired, the control unit 21 proceeds to the next step S203.
  • step S203 In step S ⁇ b> 203, the control unit 21 operates as the prediction calculation unit 212 and is based on the output value acquired from the prediction model 5, and the second allowable range defined by the second threshold 62 determined by the learning device 1. The command value to the production apparatus 3 is determined.
  • step S202 the control unit 21 acquires a correction value 73 for the reference value 70 as an output value from the prediction model 5. Therefore, the control unit 21 calculates the predicted value of the command value by correcting (for example, adding or subtracting) the reference value 70 with the acquired correction value 73. Then, the control unit 21 determines whether the calculated predicted value is within the second allowable range defined by the second threshold 62.
  • the control unit 21 determines the calculated predicted value as the command value 75.
  • the control unit 21 appropriately corrects the calculated predicted value so as to be within the second allowable range, and sets the corrected value to the command value 75. decide. For example, when the calculated predicted value exceeds the upper limit value of the second allowable range, the control unit 21 may determine the command value 75 as the upper limit value of the second allowable range. For example, when the calculated predicted value is smaller than the lower limit value of the second allowable range, the control unit 21 may determine the lower limit value of the second allowable range as the command value 75. Accordingly, when the command value 75 is determined within the second allowable range, the control unit 21 proceeds to the next step S204.
  • Step S204 In step S ⁇ b> 204, the control unit 21 operates as the operation control unit 213 and controls the operation of the production apparatus 3 based on the determined command value 75.
  • a method for controlling the operation of the production apparatus 3 based on the command value 75 may be appropriately selected according to the format of the command value.
  • the production apparatus 3 is a press machine and includes a servo driver 31 that drives the upper mold 32. Therefore, the command value 75 may indicate the number of pulses that defines the drive amount of the servo motor.
  • the control unit 21 transmits a command value 75 to the servo driver 31 of the production apparatus 3 via the external interface 24.
  • the servo driver 31 drives the servo motor based on the command value 75 received from the control device 2. Thereby, the control part 21 can control operation
  • the control unit 21 ends the process according to this operation example.
  • the format of the command value 75 may not be limited to such an example.
  • the command value 75 may be expressed by an intermediate index such as a drive amount of the servo motor and a movement amount of the upper mold 32, for example.
  • the control unit 21 may transmit the command value 75 expressed by the intermediate index as it is to the production apparatus 3, or the command value 75 expressed by the intermediate index can be used directly such as the number of pulses.
  • the converted command value 75 may be transmitted to the production apparatus 3.
  • control unit 21 ends the series of processes for controlling the operation of the production apparatus 3 according to this operation example.
  • the control unit 21 can continuously control the operation of the production apparatus 3 by repeatedly executing this series of processes.
  • the control device 2 uses the prediction model 5 for predictive control of the operation of the production device 3 (predictive control mode) and a mode for controlling the operation of the production device 3 in accordance with a user operation (manual control). Mode).
  • the control unit 21 may execute a series of processes in steps S201 to S204.
  • the control unit 21 receives the designation of the command value from the user, and controls the operation of the production apparatus 3 based on the designated command value. Good.
  • the control unit 21 may use the first threshold 60 as a constraint condition in the manual control mode. That is, the control unit 21 may accept designation of a command value within the first allowable range without accepting designation of a value exceeding the first allowable range defined by the first threshold 60.
  • step S101 the learning data set 121 for each case is collected so as to realize operation control suitable for a specific case. Therefore, according to the command value specified by the second data (correction value 1213) of the learning data set 121 for each case, the operation of the production apparatus 3 can be controlled safely. Therefore, in step S104, the second allowable range is defined so as to ensure the safety of the operation of the production apparatus 3 based on the numerical range estimated from the distribution of the second data in the learning data set 121.
  • a second threshold 62 can be determined.
  • the safety of the operation of the production apparatus 3 can be reliably ensured by determining the second threshold value 62 so as to satisfy a preset safety condition in step S104. Therefore, according to the control system 100 according to the present embodiment, it is possible to perform predictive control that can sufficiently exhibit the performance of the predictive model 5 while ensuring the safety of the operation of the production apparatus 3.
  • the first data is composed of both the feature value 1211 of the workpiece 40 and the attribute value 1212 of the environment in which the product 41 is produced. Accordingly, both the feature value 71 of the work 40 and the attribute value 72 of the environment in which the product 41 is produced are used for the input of the prediction model 5.
  • the input of the prediction model 5 may not be limited to such an example.
  • one of the feature value of the workpiece 40 and the attribute value of the environment in which the product is produced may be omitted. That is, the first data may be configured by at least one of the feature value 1211 of the workpiece 40 and the attribute value 1212 of the environment in which the product 41 is produced.
  • the prediction model 5 predicts a command value adapted to the production of the product 41 with respect to at least one input of the feature value 71 of the workpiece 40 and the attribute value 72 of the environment in which the product 41 is produced. May be built.
  • the first data may be data relating to all types of factors that can determine the operation of the target device.
  • the control device 2 may be configured to control a target device of a type other than the production device 3.
  • the prediction model 5 may be constructed to predict a command value adapted to the situation indicated by the input data with respect to the input of the same type of data as the first data.
  • 2nd data is comprised by the correction value 1213 with respect to the reference value of command value, and according to this, the prediction model 5 outputs the correction value 73 with respect to the reference value 70 of command value. It is configured.
  • the output format of the prediction model 5 may not be limited to such an example, and may be appropriately determined according to the embodiment.
  • the prediction model 5 may be configured to output the command value itself.
  • the second data may be configured by the command value itself.
  • the prediction model 5 is configured by a decision tree. However, if the configuration of the prediction model 5 can predict a command value to the target device (in the example, the production device 3 in the example) at a time point (future time point) before the time point at which the prediction process is executed, this is the case.
  • the present invention is not limited to such an example, and may be appropriately selected according to the embodiment.
  • a learning model other than a decision tree such as a neural network or a support vector machine may be used. Further, a model other than the learning model (for example, a predetermined function) may be used as the prediction model 5.
  • Control program 92 ... Storage medium, 3 ... Production equipment (target equipment), 31 ... Servo driver, 32 ... Upper die, 33 ... Lower die, 40 ... work, 41 ... product, 5 ... Prediction model (decision tree), 60 ... first threshold, 61 ... distribution, 62 ... second threshold, 70 ... Reference value (of command value), 71 ... feature amount, 72 ... attribute value, 73 ... correction value

Landscapes

  • Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Software Systems (AREA)
  • General Physics & Mathematics (AREA)
  • Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Medical Informatics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Automation & Control Theory (AREA)
  • Health & Medical Sciences (AREA)
  • Theoretical Computer Science (AREA)
  • General Engineering & Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Data Mining & Analysis (AREA)
  • Computing Systems (AREA)
  • Computational Linguistics (AREA)
  • Feedback Control In General (AREA)
  • Testing And Monitoring For Control Systems (AREA)

Abstract

A technology for performing prediction control by which performance of a prediction model can be sufficiently exerted. A control system according to an aspect of the present invention estimates a numerical value range within which a command value can fall from a distribution of second data relating to the command value in a learning data set used to construct a prediction model, and in such a manner that a first acceptable range prescribed by a preset first threshold value with respect to a command value for a subject device is extended, decides a second threshold value with respect to the command value for the subject device on the basis of the estimated numerical value range. Further, in an operational phase, on the basis of an output value from the prediction model, the control system decides a command value for the subject device within a second acceptable range prescribed by the decided second threshold value, and controls an operation of the subject device on the basis of the decided command value.

Description

制御システム、制御方法、学習装置、制御装置、学習方法及び学習プログラムControl system, control method, learning device, control device, learning method, and learning program
　本発明は、制御システム、制御方法、学習装置、制御装置、学習方法及び学習プログラムに関する。 The present invention relates to a control system, a control method, a learning device, a control device, a learning method, and a learning program.
　近年、様々な装置において、未来の状態を予測し、予測した未来の状態に適した動作の制御を行うための予測制御の技術が開発されている。例えば、特許文献１では、制御対象の複数のモデルを有し、いずれかのモデルを用いて制御量の予測値を計算する制御システムが提案されている。具体的には、この制御システムは、外部環境に応じて予測計算に用いるモデルを選択し、選択したモデルを利用して操作量を決定する。これにより、外部環境に応じた対象装置の予測制御を実現することができる。 In recent years, predictive control technology has been developed for predicting future states in various devices and controlling operations suitable for the predicted future states. For example, Patent Document 1 proposes a control system that has a plurality of models to be controlled and calculates a predicted value of a controlled variable using any one of the models. Specifically, this control system selects a model used for prediction calculation according to the external environment, and determines an operation amount using the selected model. Thereby, the prediction control of the target apparatus according to the external environment can be realized.
特開２０００－０９９１０７号公報JP 2000-099107 A
　予測モデルは、事前に収集した学習データを利用して構築される。そのため、この予測モデルによれば、学習データに表れる状況に同一又は類似するケースには、対象装置に対する指令値を適切に決定することができるが、未知のケースには、当該指令値を適切に決定できない可能性がある。換言すると、この予測モデルは、未知のケースにおいて対象装置の動作を制御する際に、可動域を超えた値、故障を発生させる値等の不適切な指令値を出力する可能性がある。そこで、予測モデルを用いる場合には、対象装置の動作の安全性を確保するために、指令値の範囲を制限する制約条件（閾値）が設けられる。例えば、特許文献１で提案されている制御システムでは、予め設定された制約条件の中で、モデルを用いて予測した制御量の予測値から最適な操作量を決定している。 The prediction model is constructed using learning data collected in advance. Therefore, according to this prediction model, it is possible to appropriately determine the command value for the target device in the case that is the same or similar to the situation that appears in the learning data, but in the unknown case, the command value is appropriately set. There is a possibility that it cannot be determined. In other words, when the operation of the target device is controlled in an unknown case, this prediction model may output an inappropriate command value such as a value that exceeds the movable range or a value that causes a failure. Therefore, when the prediction model is used, a constraint condition (threshold value) for limiting the range of the command value is provided in order to ensure the safety of the operation of the target device. For example, in the control system proposed in Patent Document 1, the optimum operation amount is determined from the predicted value of the control amount predicted using the model under the preset constraint conditions.
　しかしながら、本件発明者らは、このような予め設定された制約条件を利用する従来の制御システムでは、次のような問題点が生じ得ることを見出した。すなわち、制約条件（閾値）は、基本的には、制御システムを利用するユーザにより予め設定される。このときに、対象装置の動作の安全性を過度に考慮して、安全性を満たす範囲よりも指令値の許容範囲が狭くなるように、制約条件が設定されてしまう可能性がある。このように制約条件が設定されてしまうと、予測モデルによって決定された指令値が、安全性を満たすにも関わらず、制約条件を満たさないことで、対象装置の制御に用いる指令値として受け入れられずに、予測制御を適切に実施することができなくなってしまう。つまり、予め設定された制約条件を利用する制御システムでは、対象装置の動作の安全性を確保することはできるものの、予測モデルの性能を十分に発揮することができない可能性があるという問題点を本件発明者らは見出した。 However, the present inventors have found that the following problems may occur in a conventional control system that uses such preset constraints. That is, the constraint condition (threshold value) is basically set in advance by a user who uses the control system. At this time, considering the safety of the operation of the target device excessively, there is a possibility that the constraint condition is set so that the allowable range of the command value is narrower than the range satisfying the safety. If the constraint condition is set in this manner, the command value determined by the prediction model is accepted as the command value used for controlling the target device because the command value determined by the prediction model does not satisfy the constraint condition even though the safety is satisfied. Therefore, the predictive control cannot be properly performed. In other words, a control system that uses preset constraint conditions can ensure the safety of the operation of the target device, but may not be able to fully demonstrate the performance of the prediction model. The inventors have found out.
　本発明は、一側面では、このような実情を鑑みてなされたものであり、その目的は、予測モデルの性能を十分に発揮可能な予測制御を実施するための技術を提供することである。 In one aspect, the present invention has been made in view of such a situation, and an object thereof is to provide a technique for performing predictive control capable of sufficiently exhibiting the performance of a predictive model.
　本発明は、上述した課題を解決するために、以下の構成を採用する。 The present invention adopts the following configuration in order to solve the above-described problems.
　すなわち、本発明の一側面に係る制御システムは、対象装置の動作を決定する要因に関する第１データ、及び前記対象装置への指令値であって、前記第１データにより示される前記要因に適応した指令値に関する第２データの組み合わせによりそれぞれ構成された複数件の学習用データセットを取得する学習データ取得部と、取得した前記複数件の学習用データセットそれぞれについて、前記第１データを入力すると、前記第２データに対応する値を出力するように予測モデルを構築する学習処理部と、取得した前記複数件の学習用データセットにおける前記第２データの分布から前記指令値の取り得る数値範囲を推定する推定部と、前記対象装置への前記指令値に対して予め設定された第１の閾値により規定される第１の許容範囲を拡げるように、推定した前記数値範囲に基づいて、前記対象装置への前記指令値に対する第２の閾値を決定する閾値決定部と、運用フェーズにおいて、前記要因に関する入力データを取得する入力データ取得部と、取得した入力データを前記予測モデルに入力することで、前記予測モデルから出力値を取得し、取得した前記出力値に基づいて、決定した前記第２の閾値により規定される第２の許容範囲内で、前記対象装置に対する指令値を決定する予測演算部と、決定した前記指令値に基づいて、前記対象装置の動作を制御する動作制御部と、を備える。 That is, the control system according to one aspect of the present invention is adapted to the first data related to the factor that determines the operation of the target device and the command value to the target device, which is indicated by the first data. When the learning data acquisition unit that acquires a plurality of learning data sets each configured by a combination of second data related to the command value and each of the acquired learning data sets are input the first data, A learning processing unit that constructs a prediction model so as to output a value corresponding to the second data, and a numerical value range that the command value can take from the distribution of the second data in the acquired plurality of learning data sets. An estimation unit for estimation and a first allowable range defined by a first threshold set in advance for the command value to the target device are expanded. As described above, based on the estimated numerical range, a threshold value determination unit that determines a second threshold value for the command value to the target device, and an input data acquisition unit that acquires input data related to the factor in the operation phase; , By inputting the acquired input data to the prediction model, an output value is acquired from the prediction model, and a second allowable range defined by the second threshold value determined based on the acquired output value A prediction calculation unit that determines a command value for the target device, and an operation control unit that controls the operation of the target device based on the determined command value.
　当該構成に係る制御システムは、予測モデルの構築に利用した学習用データセットにおける第２データの分布から、対象装置への指令値の取り得る数値範囲を推定する。そして、当該構成に係る制御システムは、指令値に対して予め設定された第１の閾値により規定される第１の許容範囲を拡げるように、推定した数値範囲に基づいて、当該指令値に対する第２の閾値を決定する。当該構成に係る制御システムは、この第２の閾値により規定される第２の許容範囲を指令値の制約条件として利用する。すなわち、当該構成に係る制御システムは、予測モデルを利用する運用フェーズにおいて、この第２の閾値により規定される第２の許容範囲内で、対象装置に対する指令値を決定する。 The control system according to the configuration estimates a numerical range that can be taken by the command value to the target device from the distribution of the second data in the learning data set used for constructing the prediction model. And the control system which concerns on the said structure is the 1st with respect to the said command value based on the estimated numerical range so that the 1st tolerance | permissible_range prescribed | regulated with the 1st threshold value preset with respect to the command value may be expanded. A threshold of 2 is determined. The control system according to this configuration uses the second allowable range defined by the second threshold value as a restriction condition for the command value. That is, the control system according to the configuration determines the command value for the target device within the second allowable range defined by the second threshold value in the operation phase using the prediction model.
　これにより、安全性を過度に考慮して第１の許容範囲が狭く設定された場合であっても、この第１の許容範囲を拡げるように設定された第２の許容範囲を制約条件として利用することで、対象装置の制御に用いる指令値を許容する範囲を拡げることができる。つまり、第１の許容範囲を制約条件として利用した場合には拒絶されるような指令値の一部を対象装置の制御に用いるようにすることができる。また、各件の学習用データセットは、特定のケースに適した動作の制御を実現するように収集され得るため、各件の学習用データセットにおける第２データに基づいて指定される指令値によれば、対象装置の動作を安全に制御することができる。よって、学習用データセットにおける第２データの分布から推定される数値範囲に基づくことで、対象装置の動作の安全性を確保するように、第２の許容範囲を規定する第２の閾値を決定することができる。したがって、当該構成に係る制御システムによれば、対象装置の動作の安全性を確保しつつ、予測モデルの性能を十分に発揮可能な予測制御を実施することができる。 Thus, even when the first allowable range is set narrowly in consideration of safety excessively, the second allowable range set to expand the first allowable range is used as a constraint condition. By doing so, the range which accept | permits the command value used for control of an object apparatus can be expanded. That is, a part of the command value that is rejected when the first allowable range is used as a constraint condition can be used for controlling the target device. In addition, since each learning data set can be collected so as to realize control of operation suitable for a specific case, the command value specified based on the second data in each learning data set Accordingly, the operation of the target device can be safely controlled. Therefore, based on the numerical range estimated from the distribution of the second data in the learning data set, the second threshold value defining the second allowable range is determined so as to ensure the safety of the operation of the target device. can do. Therefore, according to the control system which concerns on the said structure, the predictive control which can fully exhibit the performance of a prediction model can be implemented, ensuring the safety | security of operation | movement of an object apparatus.
　なお、「対象装置」は、制御の対象となり得るあらゆる種類の装置を含んでもよく、例えば、ワークから製品を生産するように構成された生産装置を含んでもよい。「予測モデル」は、予測処理を実行する時点よりも先の時点（将来の時点）における生産装置への指令値を予測可能なモデルであれば特に限定されなくてもよく、実施の形態に応じて適宜選択されてよい。「予測モデル」には、例えば、決定木、ニューラルネットワーク、サポートベクタマン等の学習モデルが用いられてもよい。「第１データ」は、対象装置の動作を決定し得るあらゆる種類の要因に関するデータであってよい。「第２データ」は、対象装置への指令値を直接的に指定する値（すなわち、指令値そのもの）により構成されてもよいし、例えば、指令値の基準値に対する補正値等のように指令値を間接的に指定する値により構成されてよい。 The “target device” may include all types of devices that can be controlled, and may include, for example, a production device configured to produce a product from a workpiece. The “prediction model” is not particularly limited as long as it is a model that can predict a command value to the production apparatus at a time point (future time point) earlier than the time point at which the prediction process is executed, depending on the embodiment. May be appropriately selected. As the “prediction model”, for example, a learning model such as a decision tree, a neural network, or a support vector man may be used. The “first data” may be data relating to all types of factors that can determine the operation of the target device. The “second data” may be configured by a value that directly designates a command value to the target device (that is, the command value itself), for example, a command value such as a correction value for the reference value of the command value. You may comprise by the value which designates a value indirectly.
　上記一側面に係る制御システムにおいて、前記閾値決定部は、推定した前記数値範囲の境界値又は前記第１の閾値と当該境界値との間の値を前記第２の閾値として採用してもよい。当該構成によれば、予測モデルの性能を十分に発揮可能な予測制御を実施することができるように、第２の閾値を適切に決定することができる。 In the control system according to the above aspect, the threshold value determination unit may employ an estimated boundary value of the numerical range or a value between the first threshold value and the boundary value as the second threshold value. . According to the said structure, a 2nd threshold value can be determined appropriately so that the prediction control which can fully exhibit the performance of a prediction model can be implemented.
　上記一側面に係る制御システムにおいて、前記第１の閾値は、前記第１の許容範囲の上限値であってよく、前記閾値決定部は、前記上限値を超える値を前記第２の閾値として採用してもよい。当該構成によれば、予測モデルの性能を十分に発揮可能な予測制御を実施することができるように、第２の閾値を適切に決定することができる。 In the control system according to the above aspect, the first threshold value may be an upper limit value of the first allowable range, and the threshold value determination unit adopts a value exceeding the upper limit value as the second threshold value. May be. According to the said structure, a 2nd threshold value can be determined appropriately so that the prediction control which can fully exhibit the performance of a prediction model can be implemented.
　上記一側面に係る制御システムにおいて、前記第１の閾値は、前記第１の許容範囲の下限値であってよく、前記閾値決定部は、前記下限値より小さい値を前記第２の閾値として採用してもよい。当該構成によれば、予測モデルの性能を十分に発揮可能な予測制御を実施することができるように、第２の閾値を適切に決定することができる。 In the control system according to the above aspect, the first threshold value may be a lower limit value of the first allowable range, and the threshold value determination unit adopts a value smaller than the lower limit value as the second threshold value. May be. According to the said structure, a 2nd threshold value can be determined appropriately so that the prediction control which can fully exhibit the performance of a prediction model can be implemented.
　上記一側面に係る制御システムにおいて、前記閾値決定部は、予め設定された安全条件を満たすように前記第２の閾値を決定してもよい。当該構成によれば、対象装置の動作の安全性を確実に確保することができる。なお、「安全条件」は、適宜設定されてよく、閾値により規定されてもよいし、シミュレーション又は実機の駆動の条件により規定されてもよい。 In the control system according to the above aspect, the threshold value determination unit may determine the second threshold value so as to satisfy a preset safety condition. According to the said structure, the safety | security of operation | movement of a target apparatus can be ensured reliably. The “safety condition” may be set as appropriate, may be defined by a threshold value, or may be defined by a simulation or a condition for driving an actual machine.
　上記一側面に係る制御システムにおいて、前記第２データは、前記指令値の基準値に対する補正値により構成されてよい。当該構成によれば、予測モデルから得られる補正値を利用して、対象装置への指令値を適切に決定可能な制御システムを提供することができる。 In the control system according to the above aspect, the second data may be constituted by a correction value with respect to a reference value of the command value. According to the said structure, the control system which can determine appropriately the command value to an object apparatus can be provided using the correction value obtained from a prediction model.
　上記一側面に係る制御システムにおいて、前記対象装置は、ワークから製品を生産する生産装置であってよく、前記第１データ及び前記入力データはそれぞれ、前記ワークの特徴量及び前記製品を生産する環境の属性値の少なくとも一方により構成されてよい。当該構成によれば、予測モデルの性能を十分に発揮可能な生産装置の予測制御を実施することができる。 In the control system according to the above aspect, the target device may be a production device that produces a product from a workpiece, and the first data and the input data are an environment for producing the feature quantity of the workpiece and the product, respectively. May be constituted by at least one of the attribute values. According to the said structure, the predictive control of the production apparatus which can fully exhibit the performance of a prediction model can be implemented.
　なお、「生産装置」は、何らかの生産処理を行い、制御の対象となり得る装置であれば特に限定されなくてもよく、例えば、プレス機、射出成形機、ＮＣ旋盤、放電加工機、包装機、搬送機、検査機内の搬送機構等であってよい。「ワーク」は、生産装置の作業対象となり得る物であれば特に限定されなくてもよく、例えば、製品の原料、加工前の物、組み立て前の部品等であってよい。「製品」は、ワークに対して生産装置が生産処理を行うことで得られる物であり、最終品の他、中間品（加工途中のもの）を含んでもよい。 The “production device” is not particularly limited as long as it is a device that performs some production processing and can be controlled. For example, a press machine, an injection molding machine, an NC lathe, an electric discharge machine, a packaging machine, It may be a transport machine, a transport mechanism in an inspection machine, or the like. The “workpiece” is not particularly limited as long as it can be a work target of the production apparatus. For example, the “workpiece” may be a raw material of a product, a product before processing, a part before assembly, or the like. The “product” is a product obtained by a production apparatus performing a production process on a workpiece, and may include an intermediate product (processed product) in addition to the final product.
　「ワークの特徴量」は、ワークの何らかの特徴を示し得るものであれば特に限定されなくてもよく、実施の形態に応じて適宜選択されてよい。ワークの特徴量は、例えば、硬さ、寸法、材質、重さ、熱等を示すものであってよい。また、ワークの特徴量は、ワークの特徴を直接的に示すものであってもよいし、ワークの特徴を間接的に示すものであってもよい。ワークの特徴を直接的に示すとは、例えば、ワークの硬さ（硬度）そのものを数値、クラス等で表現することである。一方、ワークの特徴を間接的に示すとは、例えば、ワークの硬さ（硬度）を測定する際に得られた２次的指標（例えば、ワークにかかる荷重、測定の際に作用させたトルク等）を数値、クラス等で表現することである。 The “work feature amount” is not particularly limited as long as it can show some feature of the work, and may be appropriately selected according to the embodiment. The feature amount of the workpiece may indicate, for example, hardness, dimensions, material, weight, heat, and the like. Further, the feature amount of the workpiece may directly indicate the feature of the workpiece, or may indirectly indicate the feature of the workpiece. Directly indicating the characteristics of the workpiece means, for example, expressing the hardness (hardness) of the workpiece itself by a numerical value, a class, or the like. On the other hand, indirectly indicating the characteristics of the workpiece means, for example, a secondary index obtained when measuring the hardness (hardness) of the workpiece (for example, a load applied to the workpiece, a torque applied during the measurement) Etc.) are expressed by numerical values, classes, etc.
　また、「製品を生産する環境の属性値」は、生産装置が稼動する環境に関する何らかの属性を示し得るものであれば特に限定されなくてもよく、実施の形態に応じて適宜選択されてよい。製品を生産する環境の属性値は、例えば、生産装置の周囲の温度、湿度、装置の劣化度合い（例えば、経年数、加工回数等）、振動等を示すものであってよい。 Also, the “attribute value of the environment for producing the product” is not particularly limited as long as it can indicate some attribute relating to the environment in which the production apparatus operates, and may be appropriately selected according to the embodiment. The attribute value of the environment in which the product is produced may indicate, for example, the ambient temperature and humidity of the production apparatus, the degree of deterioration of the apparatus (for example, age, number of processings, etc.), vibration, and the like.
　なお、上記各形態に係る制御システムの別の態様として、本発明の一側面は、以上の各構成を実現する情報処理方法であってもよいし、プログラムであってもよいし、このようなプログラムを記憶した、コンピュータ等が読み取り可能な記憶媒体であってもよい。ここで、コンピュータ等が読み取り可能な記憶媒体とは、プログラム等の情報を、電気的、磁気的、光学的、機械的、又は、化学的作用によって蓄積する媒体である。また、上記各形態に係る制御システムの別の態様として、本発明の一側面は、以上の各構成の一部分（例えば、予測モデルを構築する部分、第２の閾値を決定する部分、予測モデル及び第２の閾値を利用する部分等）を実現する情報処理システムであってもよいし、情報処理装置であってもよいし、プログラムであってもよいし、このようなプログラムを記憶した、コンピュータ等が読み取り可能な記憶媒体であってもよい。 As another aspect of the control system according to each of the above embodiments, one aspect of the present invention may be an information processing method that realizes each of the above configurations, a program, or such It may be a computer-readable storage medium that stores the program. Here, the computer-readable storage medium is a medium that stores information such as programs by electrical, magnetic, optical, mechanical, or chemical action. Moreover, as another aspect of the control system according to each of the above aspects, one aspect of the present invention includes a part of each of the above components (for example, a part for constructing a prediction model, a part for determining a second threshold, a prediction model, and An information processing system that implements the second threshold, etc.), an information processing apparatus, a program, or a computer storing such a program May be a readable storage medium.
　例えば、本発明の一側面に係る制御方法は、コンピュータが、対象装置の動作を決定する要因に関する第１データ、及び前記対象装置への指令値であって、前記第１データにより示される前記要因に適応した指令値に関する第２データの組み合わせによりそれぞれ構成された複数件の学習用データセットを取得するステップと、取得した前記複数件の学習用データセットそれぞれについて、前記第１データを入力すると、前記第２データに対応する値を出力するように予測モデルを構築するステップと、取得した前記複数件の学習用データセットにおける前記第２データの分布から前記指令値の取り得る数値範囲を推定するステップと、前記対象装置への前記指令値に対して予め設定された第１の閾値により規定される第１の許容範囲を拡げるように、推定した前記数値範囲に基づいて、前記対象装置への前記指令値に対する第２の閾値を決定するステップと、運用フェーズにおいて、前記要因に関する入力データを取得するステップと、取得した入力データを前記予測モデルに入力することで、前記予測モデルから出力値を取得するステップと、取得した前記出力値に基づいて、決定した前記第２の閾値により規定される第２の許容範囲内で、前記対象装置に対する指令値を決定するステップと、決定した前記指令値に基づいて、前記対象装置の動作を制御するステップと、を実行する、情報処理方法である。 For example, in the control method according to one aspect of the present invention, the computer includes first data relating to a factor for determining an operation of the target device, and a command value to the target device, the factor indicated by the first data. A step of acquiring a plurality of learning data sets each constituted by a combination of second data relating to command values adapted to the step, and inputting the first data for each of the acquired learning data sets, A step of constructing a prediction model so as to output a value corresponding to the second data, and a numerical range that the command value can take is estimated from the distribution of the second data in the acquired plurality of learning data sets. A first allowable range defined by a step and a first threshold value set in advance for the command value to the target device. A step of determining a second threshold for the command value to the target device based on the estimated numerical range, a step of acquiring input data relating to the factor in an operation phase, and an acquired input By inputting data into the prediction model, an output value is obtained from the prediction model, and based on the obtained output value, within a second allowable range defined by the second threshold value determined. And an information processing method for executing a step of determining a command value for the target device and a step of controlling an operation of the target device based on the determined command value.
　また、例えば、本発明の一側面に係る学習装置は、対象装置の動作を決定する要因に関する第１データ、及び前記対象装置への指令値であって、前記第１データにより示される前記要因に適応した指令値に関する第２データの組み合わせによりそれぞれ構成された複数件の学習用データセットを取得する学習データ取得部と、取得した前記複数件の学習用データセットそれぞれについて、前記第１データを入力すると、前記第２データに対応する値を出力するように予測モデルを構築する学習処理部と、取得した前記複数件の学習用データセットにおける前記第２データの分布から前記指令値の取り得る数値範囲を推定する推定部と、前記対象装置への前記指令値に対して予め設定された第１の閾値により規定される第１の許容範囲を拡げるように、推定した前記数値範囲に基づいて、前記対象装置への前記指令値に対する第２の閾値を決定する閾値決定部と、を備える。 In addition, for example, the learning device according to one aspect of the present invention includes first data related to a factor that determines an operation of the target device, and a command value to the target device, the factor indicated by the first data A learning data acquisition unit that acquires a plurality of learning data sets each configured by a combination of second data related to the adapted command value, and the first data is input for each of the acquired learning data sets Then, a learning processing unit that builds a prediction model so as to output a value corresponding to the second data, and a numerical value that the command value can take from the distribution of the second data in the acquired plurality of learning data sets An estimation unit for estimating a range, and a first allowable range defined by a first threshold value set in advance for the command value to the target device is expanded. As provided based on the value range estimated, and a threshold determination unit that determines a second threshold value for the command value for the target device.
　また、例えば、本発明の一側面に係る制御装置は、対象装置の動作を決定する要因に関する入力データを取得する入力データ取得部と、取得した入力データを前記予測モデルに入力することで、前記予測モデルから出力値を取得し、取得した前記出力値に基づいて、上記構成に係る学習装置によって決定された前記第２の閾値により規定される第２の許容範囲内で、前記対象装置に対する指令値を決定する予測演算部と、決定した前記指令値に基づいて、前記対象装置の動作を制御する動作制御部と、を備える。 In addition, for example, the control device according to one aspect of the present invention includes an input data acquisition unit that acquires input data related to a factor that determines the operation of the target device, and the acquired input data that is input to the prediction model, An output value is obtained from the prediction model, and a command to the target device is within a second allowable range defined by the second threshold value determined by the learning device according to the configuration based on the obtained output value. A prediction calculation unit that determines a value; and an operation control unit that controls the operation of the target device based on the determined command value.
　また、例えば、本発明の一側面に係る学習方法は、コンピュータが、対象装置の動作を決定する要因に関する第１データ、及び前記対象装置への指令値であって、前記第１データにより示される前記要因に適応した指令値に関する第２データの組み合わせによりそれぞれ構成された複数件の学習用データセットを取得するステップと、取得した前記複数件の学習用データセットそれぞれについて、前記第１データを入力すると、前記第２データに対応する値を出力するように予測モデルを構築するステップと、取得した前記複数件の学習用データセットにおける前記第２データの分布から前記指令値の取り得る数値範囲を推定するステップと、前記対象装置への前記指令値に対して予め設定された第１の閾値により規定される第１の許容範囲を拡げるように、推定した前記数値範囲に基づいて、前記対象装置への前記指令値に対する第２の閾値を決定するステップと、を実行する、情報処理方法である。 Further, for example, in the learning method according to one aspect of the present invention, the computer is first data related to a factor that determines the operation of the target device, and a command value to the target device, and is indicated by the first data. The step of acquiring a plurality of learning data sets each constituted by a combination of second data relating to the command value adapted to the factor, and inputting the first data for each of the acquired learning data sets Then, a step of constructing a prediction model so as to output a value corresponding to the second data, and a numerical range that the command value can take from the distribution of the second data in the acquired plurality of learning data sets are obtained. A first allowable range defined by a step of estimating and a first threshold value set in advance for the command value to the target device. As widen the, on the basis of the value range estimated, determining a second threshold value for the command value for the target device, is executed, an information processing method.
　また、例えば、本発明の一側面に係る学習プログラムは、コンピュータに、対象装置の動作を決定する要因に関する第１データ、及び前記対象装置への指令値であって、前記第１データにより示される前記要因に適応した指令値に関する第２データの組み合わせによりそれぞれ構成された複数件の学習用データセットを取得するステップと、取得した前記複数件の学習用データセットそれぞれについて、前記第１データを入力すると、前記第２データに対応する値を出力するように予測モデルを構築するステップと、取得した前記複数件の学習用データセットにおける前記第２データの分布から前記指令値の取り得る数値範囲を推定するステップと、前記対象装置への前記指令値に対して予め設定された第１の閾値により規定される第１の許容範囲を拡げるように、推定した前記数値範囲に基づいて、前記対象装置への前記指令値に対する第２の閾値を決定するステップと、を実行させるための、プログラムである。 Further, for example, a learning program according to one aspect of the present invention is a computer that includes first data relating to factors that determine the operation of the target device, and a command value to the target device, which is indicated by the first data. The step of acquiring a plurality of learning data sets each constituted by a combination of second data relating to the command value adapted to the factor, and inputting the first data for each of the acquired learning data sets Then, a step of constructing a prediction model so as to output a value corresponding to the second data, and a numerical range that the command value can take from the distribution of the second data in the acquired plurality of learning data sets are obtained. A first step that is defined by a first threshold value that is set in advance with respect to the command value to the target device; As expanded containers range, based on the value range estimated, for executing the steps of: determining a second threshold value for the command value for the target device, a program.
　本発明によれば、予測モデルの性能を十分に発揮可能な予測制御を実施することができる。 According to the present invention, it is possible to carry out predictive control that can sufficiently exhibit the performance of the predictive model.
図１は、本発明が適用される場面の一例を模式的に例示する。FIG. 1 schematically illustrates an example of a scene to which the present invention is applied. 図２は、実施の形態に係る学習装置のハードウェア構成の一例を模式的に例示する。FIG. 2 schematically illustrates an example of a hardware configuration of the learning device according to the embodiment. 図３は、実施の形態に係る制御装置のハードウェア構成の一例を模式的に例示する。FIG. 3 schematically illustrates an example of a hardware configuration of the control device according to the embodiment. 図４は、実施の形態に係る生産装置の一例を模式的に例示する。FIG. 4 schematically illustrates an example of the production apparatus according to the embodiment. 図５Ａは、図４の生産装置における生産工程の一例を模式的に例示する。FIG. 5A schematically illustrates an example of a production process in the production apparatus of FIG. 図５Ｂは、図４の生産装置における生産工程の一例を模式的に例示する。FIG. 5B schematically illustrates an example of a production process in the production apparatus of FIG. 図５Ｃは、図４の生産装置における生産工程の一例を模式的に例示する。FIG. 5C schematically illustrates an example of a production process in the production apparatus of FIG. 図５Ｄは、図４の生産装置における生産工程の一例を模式的に例示する。FIG. 5D schematically illustrates an example of a production process in the production apparatus of FIG. 図６は、実施の形態に係る学習装置のソフトウェア構成の一例を模式的に例示する。FIG. 6 schematically illustrates an example of the software configuration of the learning device according to the embodiment. 図７Ａは、実施の形態に係る予測モデルの一例を模式的に例示する。FIG. 7A schematically illustrates an example of a prediction model according to the embodiment. 図７Ｂは、予測モデルに対する入力と出力との関係を模式的に例示する。FIG. 7B schematically illustrates the relationship between input and output for the prediction model. 図８は、実施の形態に係る制御装置のソフトウェア構成の一例を模式的に例示する。FIG. 8 schematically illustrates an example of the software configuration of the control device according to the embodiment. 図９は、実施の形態に係る学習装置の処理手順の一例を例示する。FIG. 9 illustrates an example of a processing procedure of the learning device according to the embodiment. 図１０は、指令値に関する第２データの分布の一例を模式的に例示する。FIG. 10 schematically illustrates an example of the distribution of the second data related to the command value. 図１１Ａは、第２の閾値を決定する方法の一例を模式的に例示する。FIG. 11A schematically illustrates an example of a method for determining the second threshold. 図１１Ｂは、第２の閾値を決定する方法の一例を模式的に例示する。FIG. 11B schematically illustrates an example of a method for determining the second threshold. 図１２は、実施の形態に係る制御装置の処理手順の一例を例示する。FIG. 12 illustrates an example of a processing procedure of the control device according to the embodiment.
　以下、本発明の一側面に係る実施の形態（以下、「本実施形態」とも表記する）を、図面に基づいて説明する。ただし、以下で説明する本実施形態は、あらゆる点において本発明の例示に過ぎない。本発明の範囲を逸脱することなく種々の改良や変形を行うことができることは言うまでもない。つまり、本発明の実施にあたって、実施形態に応じた具体的構成が適宜採用されてもよい。なお、本実施形態において登場するデータを自然言語により説明しているが、より具体的には、コンピュータが認識可能な疑似言語、コマンド、パラメータ、マシン語等で指定される。 Hereinafter, an embodiment according to one aspect of the present invention (hereinafter also referred to as “this embodiment”) will be described with reference to the drawings. However, this embodiment described below is only an illustration of the present invention in all respects. It goes without saying that various improvements and modifications can be made without departing from the scope of the present invention. That is, in implementing the present invention, a specific configuration according to the embodiment may be adopted as appropriate. Although data appearing in this embodiment is described in a natural language, more specifically, it is specified by a pseudo language, a command, a parameter, a machine language, or the like that can be recognized by a computer.
　§１　適用例
　まず、図１を用いて、本発明が適用される場面の一例について説明する。図１は、本実施形態に係る制御システム１００の利用場面の一例を模式的に例示する。 §1 Application Example First, an example of a scene to which the present invention is applied will be described with reference to FIG. FIG. 1 schematically illustrates an example of a usage scene of the control system 100 according to the present embodiment.
　図１で例示される制御システム１００は、ネットワークを介して接続される学習装置１及び制御装置２を備えており、生産装置３の動作を制御するように構成される。学習装置１及び制御装置２の間のネットワークの種類は、例えば、インターネット、無線通信網、移動通信網、電話網、専用網等から適宜選択されてよい。 A control system 100 illustrated in FIG. 1 includes a learning device 1 and a control device 2 connected via a network, and is configured to control the operation of the production device 3. The type of network between the learning device 1 and the control device 2 may be appropriately selected from, for example, the Internet, a wireless communication network, a mobile communication network, a telephone network, and a dedicated network.
　なお、図１の例では、学習装置１及び制御装置２は、互いに別個のコンピュータである。しかしながら、制御システム１００の構成は、このような例に限定されなくてもよい。学習装置１及び制御装置２は、一体のコンピュータで構成されてもよい。また、学習装置１及び制御装置２はそれぞれ複数台のコンピュータにより構成されてよい。 In the example of FIG. 1, the learning device 1 and the control device 2 are separate computers. However, the configuration of the control system 100 may not be limited to such an example. The learning device 1 and the control device 2 may be configured as an integrated computer. Each of the learning device 1 and the control device 2 may be composed of a plurality of computers.
　本実施形態に係る学習装置１は、生産装置３の動作を予測制御するための予測モデル（後述する予測モデル５）を構築するように構成されたコンピュータである。生産装置３は、ワークから製品を生産するよう構成されており、本発明の「対象装置」の一例である。ただし、本発明の「対象装置」は、このような生産装置３に限定されなくてもよく、制御の対象となり得るあらゆる種類の装置を含んでもよい。また、図１の例では、生産装置３は、ワークを加工するプレス機である。このプレス機は、「生産装置」の一例である。制御装置２を適用可能な生産装置は、このようなプレス機に限られなくてもよく、実施の形態に応じて適宜選択されてよい。生産装置３は、例えば、プレス機の他、射出成形機、ＮＣ旋盤、放電加工機、包装機、搬送機、検査機内の搬送機構等であってよい。 The learning device 1 according to the present embodiment is a computer configured to construct a prediction model (prediction model 5 described later) for predictive control of the operation of the production device 3. The production apparatus 3 is configured to produce a product from a workpiece, and is an example of the “target apparatus” in the present invention. However, the “target device” of the present invention is not limited to such a production device 3 and may include all types of devices that can be controlled. Moreover, in the example of FIG. 1, the production apparatus 3 is a press machine which processes a workpiece | work. This press is an example of a “production device”. The production apparatus to which the control device 2 can be applied is not limited to such a press, and may be appropriately selected according to the embodiment. The production apparatus 3 may be, for example, an injection molding machine, an NC lathe, an electric discharge machine, a packaging machine, a transport machine, a transport mechanism in an inspection machine, in addition to a press machine.
　本実施形態に係る学習装置１は、予測モデルを構築するために、複数件の学習用データセット（後述する学習用データセット１２１）を取得する。複数件の学習用データセットは、生産装置３の動作を決定する要因に関する第１データ（後述する特徴量１２１１及び属性値１２１２）、及び生産装置３への指令値であって、第１データにより示される要因に適応した指令値に関する第２データ（後述する補正値１２１３）の組み合わせによりそれぞれ構成される。学習装置１は、取得した複数件の学習用データセットそれぞれについて、第１データを入力すると、第２データに対応する値を出力するように予測モデルを構築する。 The learning device 1 according to the present embodiment acquires a plurality of learning data sets (a learning data set 121 described later) in order to construct a prediction model. The plurality of learning data sets are first data (features 1211 and attribute values 1212 described later) relating to factors that determine the operation of the production apparatus 3, and command values to the production apparatus 3, and are based on the first data. Each is constituted by a combination of second data (correction value 1213 to be described later) related to the command value adapted to the indicated factor. The learning device 1 constructs a prediction model so as to output a value corresponding to the second data when the first data is input for each of the plurality of acquired learning data sets.
　また、本実施形態に係る学習装置１は、取得した複数件の学習用データセットにおける第２データの分布から指令値の取り得る数値範囲を推定する。そして、学習装置１は、生産装置３への指令値に対して予め設定された第１の閾値により規定される第１の許容範囲を拡げるように、推定した数値範囲に基づいて、生産装置３への指令値に対する第２の閾値を決定する。 Further, the learning device 1 according to the present embodiment estimates a numerical range that the command value can take from the distribution of the second data in the acquired plurality of learning data sets. And the learning apparatus 1 is based on the estimated numerical range so that the 1st tolerance | permissible_range prescribed | regulated by the 1st threshold value preset with respect to the command value to the production apparatus 3 may be expanded. A second threshold value for the command value is determined.
　一方、本実施形態に係る制御装置２は、学習装置１により構築された予測モデルを利用して、生産装置３の動作を制御するように構成されたコンピュータである。具体的に、本実施形態に係る制御装置２は、生産装置３の動作を決定する要因に関する入力データ（後述する特徴量７１及び属性値７２）を取得する。続いて、制御装置２は、取得した入力データを予測モデルに入力することで、当該予測モデルから出力値を取得する。次に、制御装置２は、取得した出力値に基づいて、学習装置１によって決定された第２の閾値により規定される第２の許容範囲内で、生産装置３に対する指令値を決定する。そして、制御装置２は、決定した指令値に基づいて、生産装置３の動作を制御する。 On the other hand, the control device 2 according to the present embodiment is a computer configured to control the operation of the production device 3 using the prediction model constructed by the learning device 1. Specifically, the control device 2 according to the present embodiment acquires input data (a feature amount 71 and an attribute value 72 to be described later) relating to factors that determine the operation of the production device 3. Subsequently, the control device 2 acquires the output value from the prediction model by inputting the acquired input data to the prediction model. Next, the control device 2 determines a command value for the production device 3 within a second allowable range defined by the second threshold value determined by the learning device 1 based on the acquired output value. And the control apparatus 2 controls operation | movement of the production apparatus 3 based on the determined command value.
　以上のとおり、本実施形態に係る制御システム１００では、予め設定された第１の閾値により規定される第１の許容範囲ではなく、第１の許容範囲を拡げるように設定された第２の閾値により規定される第２の許容範囲が指令値の制約条件として利用される。これにより、安全性を過度に考慮して第１の許容範囲が狭く設定された場合であっても、生産装置３の動作の制御に用いる指令値を許容する範囲を拡げることができる。つまり、第１の許容範囲を制約条件として利用した場合には拒絶されるような指令値の一部を生産装置３の動作の制御に用いるようにすることができる。 As described above, in the control system 100 according to the present embodiment, the second threshold value set to expand the first allowable range instead of the first allowable range defined by the preset first threshold value. The second allowable range defined by is used as a constraint value for the command value. Thereby, even if it is a case where the 1st permissible range is set narrowly in consideration of safety too much, the range which accepts the command value used for control of operation of production device 3 can be expanded. That is, a part of the command value that is rejected when the first allowable range is used as a constraint condition can be used for controlling the operation of the production apparatus 3.
　更に、各件の学習用データセットは、特定のケースに適した動作の制御を実現するように収集される。そのため、各件の学習用データセットにおける第２データに基づいて指定される指令値によれば、生産装置３の動作を安全に制御することができる。よって、学習用データセットにおける第２データの分布から推定される数値範囲に基づくことで、生産装置３の動作の安全性を確保するように、第２の許容範囲を規定する第２の閾値を決定することができる。したがって、本実施形態に係る制御システム１００によれば、生産装置３の動作の安全性を確保しつつ、予測モデルの性能を十分に発揮可能な予測制御を実施することができる。 Furthermore, each learning data set is collected so as to realize operation control suitable for a specific case. Therefore, according to the command value specified based on the second data in each learning data set, the operation of the production apparatus 3 can be controlled safely. Therefore, based on the numerical range estimated from the distribution of the second data in the learning data set, the second threshold value defining the second allowable range is set so as to ensure the safety of the operation of the production apparatus 3. Can be determined. Therefore, according to the control system 100 according to the present embodiment, it is possible to perform predictive control that can sufficiently exhibit the performance of the predictive model while ensuring the safety of the operation of the production apparatus 3.
　§２　構成例
　［ハードウェア構成］
　＜学習装置＞
　次に、図２を用いて、本実施形態に係る学習装置１のハードウェア構成の一例について説明する。図２は、本実施形態に係る学習装置１のハードウェア構成の一例を模式的に例示する。 §2 Configuration example [Hardware configuration]
<Learning device>
Next, an example of the hardware configuration of the learning device 1 according to the present embodiment will be described with reference to FIG. FIG. 2 schematically illustrates an example of a hardware configuration of the learning device 1 according to the present embodiment.
　図２に示されるとおり、本実施形態に係る学習装置１は、制御部１１、記憶部１２、通信インタフェース１３、入力装置１４、出力装置１５、及びドライブ１６が電気的に接続されたコンピュータである。なお、図２では、通信インタフェースを「通信Ｉ／Ｆ」と記載している。 As illustrated in FIG. 2, the learning device 1 according to the present embodiment is a computer in which a control unit 11, a storage unit 12, a communication interface 13, an input device 14, an output device 15, and a drive 16 are electrically connected. . In FIG. 2, the communication interface is described as “communication I / F”.
　制御部１１は、ハードウェアプロセッサであるＣＰＵ（Central Processing Unit）、ＲＡＭ（Random Access Memory）、ＲＯＭ（Read Only Memory）等を含み、プログラム及び各種データに基づいて情報処理を実行するように構成される。記憶部１２は、メモリの一例であり、例えば、ハードディスクドライブ、ソリッドステートドライブ等で構成される。本実施形態では、記憶部１２は、制御部１１（ＣＰＵ）により実行される学習プログラム８１、複数件の学習用データセット１２１、学習結果データ１２５等の各種情報を記憶する。 The control unit 11 includes a CPU (Central Processing Unit), a RAM (Random Access Memory), a ROM (Read Only Memory), etc., which are hardware processors, and is configured to execute information processing based on programs and various data. The The storage unit 12 is an example of a memory, and includes, for example, a hard disk drive or a solid state drive. In the present embodiment, the storage unit 12 stores various information such as a learning program 81 executed by the control unit 11 (CPU), a plurality of learning data sets 121, and learning result data 125.
　学習プログラム８１は、予測モデルを構築する機械学習の後述する情報処理（図９）を学習装置１に実行させ、当該機械学習の結果として学習結果データ１２５を生成するためのプログラムである。学習プログラム８１は、当該情報処理の一連の命令を含む。複数件の学習用データセット１２１は、生産装置３による製品の生産に適応した指令値を予測する能力を獲得した予測モデルを構築するための機械学習に利用されるデータである。詳細は後述する。 The learning program 81 is a program for causing the learning apparatus 1 to execute later-described information processing (FIG. 9) of machine learning for constructing a prediction model and generating learning result data 125 as a result of the machine learning. The learning program 81 includes a series of instructions for the information processing. The plurality of learning data sets 121 are data used for machine learning for constructing a prediction model that has acquired the ability to predict a command value adapted to product production by the production apparatus 3. Details will be described later.
　通信インタフェース１３は、例えば、有線ＬＡＮ（Local Area Network）モジュール、無線ＬＡＮモジュール等であり、ネットワークを介した有線又は無線通信を行うためのインタフェースである。学習装置１は、この通信インタフェース１３を利用することで、ネットワークを介したデータ通信を他の情報処理装置（例えば、制御装置２）と行うことができる。また、学習装置１は、この通信インタフェース１３を利用することで、生成した学習結果データ１２５を外部の装置に配信することができる。 The communication interface 13 is, for example, a wired LAN (Local Area Network) module, a wireless LAN module, or the like, and is an interface for performing wired or wireless communication via a network. The learning device 1 can perform data communication via the network with another information processing device (for example, the control device 2) by using the communication interface 13. In addition, the learning device 1 can distribute the generated learning result data 125 to an external device by using the communication interface 13.
　入力装置１４は、例えば、マウス、キーボード等の入力を行うための装置である。また、出力装置１５は、例えば、ディスプレイ、スピーカ等の出力を行うための装置である。オペレータは、入力装置１４及び出力装置１５を利用することで、学習装置１を操作することができる。 The input device 14 is a device for inputting, for example, a mouse and a keyboard. The output device 15 is a device for outputting, for example, a display or a speaker. The operator can operate the learning device 1 by using the input device 14 and the output device 15.
　ドライブ１６は、例えば、ＣＤドライブ、ＤＶＤドライブ等であり、記憶媒体９１に記憶されたプログラムを読み込むためのドライブ装置である。ドライブ１６の種類は、記憶媒体９１の種類に応じて適宜選択されてよい。上記学習プログラム８１及び学習用データセット１２１の少なくとも一方は、この記憶媒体９１に記憶されていてもよい。 The drive 16 is, for example, a CD drive, a DVD drive, or the like, and is a drive device for reading a program stored in the storage medium 91. The type of the drive 16 may be appropriately selected according to the type of the storage medium 91. At least one of the learning program 81 and the learning data set 121 may be stored in the storage medium 91.
　記憶媒体９１は、コンピュータその他装置、機械等が、記録されたプログラム等の情報を読み取り可能なように、当該プログラム等の情報を、電気的、磁気的、光学的、機械的又は化学的作用によって蓄積する媒体である。学習装置１は、この記憶媒体９１から、上記学習プログラム８１及び学習用データセット１２１の少なくとも一方を取得してもよい。 The storage medium 91 stores information such as a program by electrical, magnetic, optical, mechanical, or chemical action so that a computer or other device, machine, or the like can read the recorded program or the like. It is a medium to accumulate. The learning device 1 may acquire at least one of the learning program 81 and the learning data set 121 from the storage medium 91.
　ここで、図２では、記憶媒体９１の一例として、ＣＤ、ＤＶＤ等のディスク型の記憶媒体を例示している。しかしながら、記憶媒体９１の種類は、ディスク型に限定される訳ではなく、ディスク型以外であってもよい。ディスク型以外の記憶媒体として、例えば、フラッシュメモリ等の半導体メモリを挙げることができる。 Here, in FIG. 2, as an example of the storage medium 91, a disk-type storage medium such as a CD or a DVD is illustrated. However, the type of the storage medium 91 is not limited to the disk type and may be other than the disk type. Examples of the storage medium other than the disk type include a semiconductor memory such as a flash memory.
　なお、学習装置１の具体的なハードウェア構成に関して、実施形態に応じて、適宜、構成要素の省略、置換及び追加が可能である。例えば、制御部１１は、複数のハードウェアプロセッサを含んでもよい。ハードウェアプロセッサは、マイクロプロセッサ、ＦＰＧＡ（field-programmable gate array）等で構成されてよい。記憶部１２は、制御部１１に含まれるＲＡＭ及びＲＯＭにより構成されてもよい。通信インタフェース１３、入力装置１４、出力装置１５及びドライブ１６の少なくともいずれかは省略されてもよい。学習装置１は、複数台の情報処理装置で構成されてもよい。この場合、各コンピュータのハードウェア構成は、一致していてもよいし、一致していなくてもよい。また、学習装置１には、提供されるサービス専用に設計された情報処理装置の他、汎用のサーバ装置、汎用のＰＣ（Personal Computer）等が用いられてもよい。 It should be noted that regarding the specific hardware configuration of the learning device 1, the components can be omitted, replaced, and added as appropriate according to the embodiment. For example, the control unit 11 may include a plurality of hardware processors. The hardware processor may be configured by a microprocessor, an FPGA (field-programmable gate array), or the like. The storage unit 12 may be configured by a RAM and a ROM included in the control unit 11. At least one of the communication interface 13, the input device 14, the output device 15, and the drive 16 may be omitted. The learning device 1 may be composed of a plurality of information processing devices. In this case, the hardware configurations of the computers may or may not match. The learning device 1 may be a general-purpose server device, a general-purpose PC (Personal Computer), or the like, in addition to an information processing device designed exclusively for the provided service.
　＜制御装置＞
　次に、図３を用いて、本実施形態に係る制御装置２のハードウェア構成の一例について説明する。図３は、本実施形態に係る制御装置２のハードウェア構成の一例を模式的に例示する。 <Control device>
Next, an example of the hardware configuration of the control device 2 according to the present embodiment will be described with reference to FIG. FIG. 3 schematically illustrates an example of a hardware configuration of the control device 2 according to the present embodiment.
　図３に示されるとおり、本実施形態に係る制御装置２は、制御部２１、記憶部２２、通信インタフェース２３、外部インタフェース２４、入力装置２５、出力装置２６、及びドライブ２７が電気的に接続されたコンピュータである。なお、図３では、通信インタフェース及び外部インタフェースをそれぞれ「通信Ｉ／Ｆ」及び「外部Ｉ／Ｆ」と記載している。 As shown in FIG. 3, the control device 2 according to this embodiment includes a control unit 21, a storage unit 22, a communication interface 23, an external interface 24, an input device 25, an output device 26, and a drive 27 that are electrically connected. Computer. In FIG. 3, the communication interface and the external interface are described as “communication I / F” and “external I / F”, respectively.
　制御部２１は、上記制御部１１と同様に、ハードウェアプロセッサであるＣＰＵ、ＲＡＭ、ＲＯＭ等を含み、プログラム及び各種データに基づいて情報処理を実行するように構成される。記憶部２２は、例えば、ハードディスクドライブ、ソリッドステートドライブ等で構成される。記憶部２２は、制御部２１（ＣＰＵ）により実行される制御プログラム８２、学習結果データ１２５等の各種情報を記憶する。 Similarly to the control unit 11, the control unit 21 includes a CPU, RAM, ROM, and the like, which are hardware processors, and is configured to execute information processing based on programs and various data. The storage unit 22 is configured by, for example, a hard disk drive, a solid state drive, or the like. The storage unit 22 stores various information such as a control program 82 executed by the control unit 21 (CPU) and learning result data 125.
　制御プログラム８２は、生産装置３の動作を制御する後述の情報処理（図１２）を制御装置２に実行させるためのプログラムであり、当該情報処理の一連の命令を含む。学習結果データ１２５は、学習済みの予測モデルの設定を行うためのデータである。詳細は後述する。 The control program 82 is a program for causing the control device 2 to execute information processing (FIG. 12) described later for controlling the operation of the production device 3, and includes a series of instructions for the information processing. The learning result data 125 is data for setting a learned prediction model. Details will be described later.
　通信インタフェース２３は、例えば、有線ＬＡＮモジュール、無線ＬＡＮモジュール等であり、ネットワークを介した有線又は無線通信を行うためのインタフェースである。制御装置２は、この通信インタフェース２３を利用することで、ネットワークを介したデータ通信を他の情報処理装置（例えば、学習装置１）と行うことができる。 The communication interface 23 is, for example, a wired LAN module, a wireless LAN module, or the like, and is an interface for performing wired or wireless communication via a network. By using this communication interface 23, the control device 2 can perform data communication via the network with another information processing device (for example, the learning device 1).
　外部インタフェース２４は、例えば、ＵＳＢ（Universal Serial Bus）ポート、専用ポート等であり、外部装置と接続するためのインタフェースである。外部インタフェース２４の種類及び数は、接続される外部装置の種類及び数に応じて適宜選択されてよい。本実施形態では、制御装置２は、外部インタフェース２４を介して、生産装置３に接続される。これにより、制御装置２は、生産装置３に対して指令値を送信することで、生産装置３の動作を制御することができる。 The external interface 24 is, for example, a USB (Universal Serial Bus) port, a dedicated port, or the like, and is an interface for connecting to an external device. The type and number of external interfaces 24 may be appropriately selected according to the type and number of external devices to be connected. In the present embodiment, the control device 2 is connected to the production device 3 via the external interface 24. Thereby, the control apparatus 2 can control the operation of the production apparatus 3 by transmitting a command value to the production apparatus 3.
　入力装置２５は、例えば、マウス、キーボード等の入力を行うための装置である。また、出力装置２６は、例えば、ディスプレイ、スピーカ等の出力を行うための装置である。オペレータは、入力装置２５及び出力装置２６を介して、制御装置２を操作することができる。 The input device 25 is a device for inputting, for example, a mouse and a keyboard. The output device 26 is a device for outputting, for example, a display or a speaker. An operator can operate the control device 2 via the input device 25 and the output device 26.
　ドライブ２７は、例えば、ＣＤドライブ、ＤＶＤドライブ等であり、記憶媒体９２に記憶されたプログラムを読み込むためのドライブ装置である。ドライブ２７の種類は、記憶媒体９２の種類に応じて適宜選択されてよい。上記制御プログラム８２及び学習結果データ１２５の少なくとも一方は、この記憶媒体９２に記憶されていてもよい。 The drive 27 is, for example, a CD drive, a DVD drive, or the like, and is a drive device for reading a program stored in the storage medium 92. The type of the drive 27 may be appropriately selected according to the type of the storage medium 92. At least one of the control program 82 and the learning result data 125 may be stored in the storage medium 92.
　記憶媒体９２は、コンピュータその他装置、機械等が記録されたプログラム等の情報を読み取り可能なように、当該プログラム等の情報を、電気的、磁気的、光学的、機械的又は化学的作用によって蓄積する媒体である。制御装置２は、この記憶媒体９２から、上記制御プログラム８２及び学習結果データ１２５の少なくとも一方を取得してもよい。 The storage medium 92 stores information such as a program by an electrical, magnetic, optical, mechanical, or chemical action so that information such as a program recorded by a computer or other device or machine can be read. It is a medium to do. The control device 2 may acquire at least one of the control program 82 and the learning result data 125 from the storage medium 92.
　ここで、図３では、上記図２と同様に、記憶媒体９２の一例として、ＣＤ、ＤＶＤ等のディスク型の記憶媒体を例示している。しかしながら、記憶媒体９２の種類は、ディスク型に限定される訳ではなく、ディスク型以外であってもよい。ディスク型以外の記憶媒体として、例えば、フラッシュメモリ等の半導体メモリを挙げることができる。 Here, FIG. 3 illustrates a disk-type storage medium such as a CD and a DVD as an example of the storage medium 92 as in FIG. However, the type of the storage medium 92 is not limited to the disk type and may be other than the disk type. Examples of the storage medium other than the disk type include a semiconductor memory such as a flash memory.
　なお、制御装置２の具体的なハードウェア構成に関して、実施形態に応じて、適宜、構成要素の省略、置換及び追加が可能である。例えば、制御部２１は、複数のハードウェアプロセッサを含んでもよい。ハードウェアプロセッサは、マイクロプロセッサ、ＦＰＧＡ、ＤＳＰ等で構成されてよい。記憶部２２は、制御部２１に含まれるＲＡＭ及びＲＯＭにより構成されてもよい。通信インタフェース２３、外部インタフェース２４、入力装置２５、出力装置２６及びドライブ２７の少なくともいずれかは省略されてもよい。制御装置２は、複数台のコンピュータで構成されてもよい。この場合、各コンピュータのハードウェア構成は、一致していてもよいし、一致していなくてもよい。また、制御装置２は、提供されるサービス専用に設計された情報処理装置の他、汎用のコントローラ、汎用のサーバ装置、汎用のデスクトップＰＣ、ノートＰＣ、タブレットＰＣ等であってもよい。 It should be noted that regarding the specific hardware configuration of the control device 2, the components can be omitted, replaced, and added as appropriate according to the embodiment. For example, the control unit 21 may include a plurality of hardware processors. The hardware processor may be configured by a microprocessor, FPGA, DSP, or the like. The storage unit 22 may be configured by a RAM and a ROM included in the control unit 21. At least one of the communication interface 23, the external interface 24, the input device 25, the output device 26, and the drive 27 may be omitted. The control device 2 may be composed of a plurality of computers. In this case, the hardware configurations of the computers may or may not match. The control device 2 may be a general-purpose controller, a general-purpose server device, a general-purpose desktop PC, a notebook PC, a tablet PC, or the like, in addition to an information processing device designed exclusively for the provided service.
　＜生産装置＞
　次に、図４を用いて、本実施形態に係る生産装置３のハードウェア構成の一例について説明する。図４は、本実施形態に係る生産装置３のハードウェア構成の一例を模式的に例示する。 <Production equipment>
Next, an example of the hardware configuration of the production apparatus 3 according to this embodiment will be described with reference to FIG. FIG. 4 schematically illustrates an example of a hardware configuration of the production apparatus 3 according to the present embodiment.
　本実施形態に係る生産装置３は、サーボドライバ３１、上側金型３２、及び下側金型３３を備えている。下側金型３３が固定されているのに対して、上側金型３２は、サーボモータ（不図示）によって、上下方向に移動可能に構成されている。これにより、上側金型３２は、下側金型３３にワークを押し付けて、ワークの成形を行ったり、下側金型３３から離れたりすることができる。サーボドライバ３１は、制御装置２からの指令値に基づいて、上側金型３２のサーボモータを駆動するように構成される。 The production apparatus 3 according to the present embodiment includes a servo driver 31, an upper mold 32, and a lower mold 33. While the lower mold 33 is fixed, the upper mold 32 is configured to be movable in the vertical direction by a servo motor (not shown). As a result, the upper mold 32 can press the workpiece against the lower mold 33 to mold the workpiece, or to move away from the lower mold 33. The servo driver 31 is configured to drive the servo motor of the upper mold 32 based on the command value from the control device 2.
　次に、図５Ａ～図５Ｄを用いて、生産装置３における生産工程の一例を模式的に例示する。生産装置３は、例えば、生産ラインに配置される。図５Ａに示されるとおり、初期状態では、上側金型３２は、下側金型３３から離れた待機位置に配置され、下側金型３３にワーク４０が搬送されるまで待機する。ワーク４０は、例えば、金属製の板材である。ただし、ワーク４０は、このような例に限定される訳ではなく、生産装置３の種類に応じて適宜選択されてよい。ワーク４０は、例えば、製品の原料、加工前の物、組み立て前の部品等であってよい。 Next, an example of the production process in the production apparatus 3 will be schematically illustrated with reference to FIGS. 5A to 5D. The production apparatus 3 is arranged on a production line, for example. As shown in FIG. 5A, in the initial state, the upper mold 32 is arranged at a standby position away from the lower mold 33 and waits until the workpiece 40 is conveyed to the lower mold 33. The workpiece 40 is, for example, a metal plate material. However, the workpiece 40 is not limited to such an example, and may be appropriately selected according to the type of the production apparatus 3. The workpiece 40 may be, for example, a raw material of a product, a product before processing, a part before assembly, or the like.
　下側金型３３の所定の位置にワーク４０が配置された後、生産装置３は、図５Ｂに示されるとおり、サーボドライバ３１により上側金型３２のサーボモータを駆動し、上側金型３２を成形開始位置に配置する。成形開始位置は、例えば、上側金型３２の先端がワーク４０に接触する又はその直前の位置である。 After the workpiece 40 is arranged at a predetermined position of the lower mold 33, the production apparatus 3 drives the servo motor of the upper mold 32 by the servo driver 31, as shown in FIG. Place at the molding start position. The molding start position is, for example, a position where the tip of the upper mold 32 comes into contact with the workpiece 40 or immediately before it.
　そして、生産装置３は、図５Ｃに示されるとおり、サーボドライバ３１により上側金型３２のサーボモータを更に駆動し、上側金型３２を目標位置（下死点）まで移動させ、上側金型３２及び下側金型３３によりワーク４０の成形を行う。これにより、生産装置３は、ワーク４０から製品４１を生産することができる。なお、この製品４１は、ワーク４０に対して生産装置３が生産処理を行うことで得られる物であれば特に限定されなくてもよく、最終品であってもよいし、中間品（加工途中のもの）であってもよい。 Then, as shown in FIG. 5C, the production apparatus 3 further drives the servo motor of the upper mold 32 by the servo driver 31, moves the upper mold 32 to the target position (bottom dead center), and moves the upper mold 32. Then, the workpiece 40 is formed by the lower mold 33. Thereby, the production apparatus 3 can produce the product 41 from the workpiece 40. The product 41 is not particularly limited as long as it is a product obtained by the production apparatus 3 performing a production process on the workpiece 40, and may be a final product or an intermediate product (during processing). May be).
　成形が完了した後、生産装置３は、図５Ｄに示されるとおり、サーボドライバ３１により上側金型３２のサーボモータを駆動し、上側金型３２を待機位置まで移動させる。そして、ワーク４０を成形することで得られた製品４１をベルトコンベア（不図示）等により生産装置３から搬送する。これにより、ワーク４０から製品４１を生産する一連の生産工程が完了する。 After the molding is completed, the production apparatus 3 drives the servo motor of the upper mold 32 by the servo driver 31, as shown in FIG. 5D, and moves the upper mold 32 to the standby position. And the product 41 obtained by shape | molding the workpiece | work 40 is conveyed from the production apparatus 3 by a belt conveyor (not shown). Thus, a series of production steps for producing the product 41 from the workpiece 40 is completed.
　この生産工程において、図５Ｃにおけるプレス時間が不十分であったり、上側金型３２が下死点に到達するまでサーボモータを駆動していなかったりすると、得られる製品４１の品質が悪化してしまう。そこで、従来、現場の作業者が、定期的に製品の品質をチェックし、生産装置の動作の設定を調節することで、不良品の発生を抑制していた。これに対して、本実施形態に係る制御装置２は、予測モデルを利用することで、生産工程に不良が生じないように、生産装置３への適切な指令値を予測する。これにより、制御装置２は、不良品の発生を抑制するように、生産装置３の動作を自動的に調節する。 In this production process, if the press time in FIG. 5C is insufficient, or if the servo motor is not driven until the upper die 32 reaches the bottom dead center, the quality of the product 41 obtained will deteriorate. . Therefore, conventionally, on-site workers have regularly checked the quality of the products and adjusted the operation settings of the production apparatus to suppress the occurrence of defective products. On the other hand, the control device 2 according to the present embodiment predicts an appropriate command value to the production device 3 by using the prediction model so that no defect occurs in the production process. Thereby, the control apparatus 2 adjusts operation | movement of the production apparatus 3 automatically so that generation | occurrence | production of inferior goods may be suppressed.
　［ソフトウェア構成］
　＜学習装置＞
　次に、図６を用いて、本実施形態に係る学習装置１のソフトウェア構成の一例について説明する。図６は、本実施形態に係る学習装置１のソフトウェア構成の一例を模式的に例示する。 Software configuration
<Learning device>
Next, an example of the software configuration of the learning device 1 according to the present embodiment will be described with reference to FIG. FIG. 6 schematically illustrates an example of the software configuration of the learning device 1 according to the present embodiment.
　学習装置１の制御部１１は、記憶部１２に記憶された学習プログラム８１をＲＡＭに展開する。そして、制御部１１は、ＲＡＭに展開された学習プログラム８１をＣＰＵにより解釈及び実行して、学習プログラム８１に含まれる一連の命令に基づいて、各構成要素を制御する。これによって、図６に示されるとおり、本実施形態に係る学習装置１は、学習データ取得部１１１、学習処理部１１２、推定部１１３、及び閾値決定部１１４をソフトウェアモジュールとして備えるコンピュータとして動作する。すなわち、本実施形態では、各ソフトウェアモジュールは、制御部１１（ＣＰＵ）により実現される。 The control unit 11 of the learning device 1 expands the learning program 81 stored in the storage unit 12 in the RAM. Then, the control unit 11 interprets and executes the learning program 81 expanded in the RAM, and controls each component based on a series of instructions included in the learning program 81. Accordingly, as illustrated in FIG. 6, the learning device 1 according to the present embodiment operates as a computer including the learning data acquisition unit 111, the learning processing unit 112, the estimation unit 113, and the threshold value determination unit 114 as software modules. That is, in the present embodiment, each software module is realized by the control unit 11 (CPU).
　学習データ取得部１１１は、予測モデル５の機械学習に利用する複数件の学習用データセット１２１を取得する。各件の学習用データセット１２１は、生産装置３の動作を決定する要因に関する第１データ、及び生産装置３への指令値であって、第１データにより示される要因に適応した指令値に関する第２データの組み合わせで構成される。具体的に、本実施形態では、第１データは、ワーク４０の特徴量１２１１及び製品４１を生産する環境の属性値１２１２により構成される。また、第２データは、指令値の基準値に対する補正値１２１３であって、特徴量１２１１及び属性値１２１２により示される状況に適応した指令値が得られるように決定された補正値１２１３により構成される。第１データは、訓練データ（入力データ）に対応し、第２データは、教師データ（正解データ）に対応する。 The learning data acquisition unit 111 acquires a plurality of learning data sets 121 used for machine learning of the prediction model 5. Each learning data set 121 includes first data relating to factors that determine the operation of the production apparatus 3 and command values to the production apparatus 3, which are command values adapted to the factors indicated by the first data. It consists of a combination of two data. Specifically, in the present embodiment, the first data is composed of the feature value 1211 of the work 40 and the attribute value 1212 of the environment in which the product 41 is produced. The second data is a correction value 1213 with respect to the reference value of the command value, and is composed of a correction value 1213 determined so that a command value adapted to the situation indicated by the feature value 1211 and the attribute value 1212 can be obtained. The The first data corresponds to training data (input data), and the second data corresponds to teacher data (correct answer data).
　学習処理部１１２は、取得した複数件の学習用データセット１２１を利用した機械学習を行うことにより、学習済みの予測モデル５を構築する。すなわち、学習処理部１１２は、取得した各件の学習用データセット１２１について、第１データ（特徴量１２１１及び属性値１２１２）を入力すると、入力した第１データに関連付けられた第２データ（補正値１２１３）に対応する値を出力するように予測モデル５を構築する。そして、学習処理部１１２は、構築した学習済みの予測モデル５に関する情報を学習結果データ１２５として記憶部１２に格納する。 The learning processing unit 112 constructs the learned prediction model 5 by performing machine learning using the plurality of acquired learning data sets 121. That is, when the learning processing unit 112 inputs the first data (feature value 1211 and attribute value 1212) for each acquired learning data set 121, the second data (correction) associated with the input first data is input. The prediction model 5 is constructed so as to output a value corresponding to the value 1213). Then, the learning processing unit 112 stores information on the constructed learned prediction model 5 in the storage unit 12 as learning result data 125.
　推定部１１３は、取得した複数件の学習用データセット１２１における第２データ（補正値１２１３）の分布６１から指令値の取り得る数値範囲を推定する。そして、閾値決定部１１４は、生産装置３への指令値に対して予め設定された第１の閾値６０により規定される第１の許容範囲を拡げるように、推定した数値範囲に基づいて、生産装置３への指令値に対する第２の閾値６２を決定する。 The estimation unit 113 estimates a possible numerical range of the command value from the distribution 61 of the second data (correction value 1213) in the acquired plurality of learning data sets 121. Then, the threshold value determination unit 114 performs the production based on the estimated numerical range so as to expand the first allowable range defined by the first threshold value 60 set in advance with respect to the command value to the production apparatus 3. A second threshold 62 for the command value to the device 3 is determined.
　（予測モデル）
　次に、図７Ａ及び図７Ｂを更に用いて、本実施形態に係る予測モデル５について説明する。図７Ａは、本実施形態に係る予測モデル５の構成の一例を模式的に例示する。また、図７Ｂは、予測モデル５に対する入力と出力との関係を模式的に例示する。 (Prediction model)
Next, the prediction model 5 according to the present embodiment will be described with reference to FIGS. 7A and 7B. FIG. 7A schematically illustrates an example of the configuration of the prediction model 5 according to the present embodiment. FIG. 7B schematically illustrates the relationship between the input and output for the prediction model 5.
　図７Ａに示されるとおり、本実施形態に係る予測モデル５は、決定木（具体的には、回帰木）によって構成されている。予測モデル５（決定木）は、根ノードＲ、葉ノードＬ１～Ｌ５、及び根ノードＲと葉ノードＬ１～Ｌ５の間に配置される中間ノードＮ１～Ｎ３を含んでいる。各ノードの間にはリンクが設けられる。図７Ａの例では、根ノードＲと中間ノード（Ｎ１、Ｎ２）との間、中間ノードＮ１と各葉ノード（Ｌ１、Ｌ２）との間、中間ノードＮ２と葉ノードＬ３及び中間ノードＮ３との間、中間ノードＮ３と各葉ノード（Ｌ４、Ｌ５）との間にそれぞれ、リンクが設けられている。 As shown in FIG. 7A, the prediction model 5 according to the present embodiment is configured by a decision tree (specifically, a regression tree). The prediction model 5 (decision tree) includes a root node R, leaf nodes L1 to L5, and intermediate nodes N1 to N3 arranged between the root node R and the leaf nodes L1 to L5. A link is provided between each node. In the example of FIG. 7A, between the root node R and the intermediate nodes (N1, N2), between the intermediate node N1 and each leaf node (L1, L2), between the intermediate node N2, the leaf node L3, and the intermediate node N3. A link is provided between the intermediate node N3 and each leaf node (L4, L5).
　なお、図７Ａの例では、決定木の深さは４であり、中間ノードの数は３つであり、葉ノードの数は５つである。しかしながら、決定木の深さ、中間ノードの数、及び葉ノードの数は、このような例に限定されなくてもよく、実施の形態に応じて適宜決定されてよい。また、図７Ａの例では、根ノードＲから各葉ノードＬ１～Ｌ５にリンクが設けられていない。しかしながら、決定木の構成は、このような例に限定されなくてもよく、根ノードからのリンクに接続される葉ノードが存在してもよい。 In the example of FIG. 7A, the depth of the decision tree is 4, the number of intermediate nodes is 3, and the number of leaf nodes is 5. However, the depth of the decision tree, the number of intermediate nodes, and the number of leaf nodes do not have to be limited to such an example, and may be appropriately determined according to the embodiment. In the example of FIG. 7A, no link is provided from the root node R to each of the leaf nodes L1 to L5. However, the configuration of the decision tree may not be limited to such an example, and there may be leaf nodes connected to the link from the root node.
　このような予測モデル５の演算処理は、決定木の根ノードＲから葉ノードＬ１～Ｌ５に向けてリンクをたどる探索処理である。すなわち、根ノードＲから葉ノードＬ１～Ｌ５の経路（図７Ａの例では、根ノードＲ及び中間ノードＮ１～Ｎ３）には、分岐条件が紐付けられている。図７Ａの例では、根ノードＲには「ｘ０＜２５００」の分岐条件が、中間ノードＮ１には「ｘ１＜２０」の分岐条件が、中間ノードＮ２には「ｘ１＜３５」の分岐条件が、中間ノードＮ３には「ｘ０＜３５００」の分岐条件が紐付けられている。一方、各葉ノードＬ１～Ｌ５には、図７Ｂに示されるとおり、予測モデル５の演算処理の最終結果（クラスＣ１～Ｃ５）が紐付けられている。 The calculation process of the prediction model 5 is a search process that follows links from the root node R of the decision tree toward the leaf nodes L1 to L5. That is, a branch condition is associated with the route from the root node R to the leaf nodes L1 to L5 (in the example of FIG. 7A, the root node R and the intermediate nodes N1 to N3). In the example of FIG. 7A, the root node R has a branch condition “x0 <2500”, the intermediate node N1 has a branch condition “x1 <20”, and the intermediate node N2 has a branch condition “x1 <35”. The branch condition “x0 <3500” is associated with the intermediate node N3. On the other hand, as shown in FIG. 7B, the final results (classes C1 to C5) of the calculation process of the prediction model 5 are associated with the leaf nodes L1 to L5.
　本実施形態では、各葉ノードＬ１～Ｌ５（クラスＣ１～Ｃ５）には、入力される特徴量及び属性値に応じた補正値が紐付けられる。すなわち、学習処理部１１２は、特徴量１２１１及び属性値１２１２を入力すると、入力した特徴量１２１１及び属性値１２１２に関連付けられた補正値１２１３に対応するクラスの葉ノードに到達するように予測モデル５（決定木）を構築する。そして、学習処理部１１２は、構築した学習済みの予測モデル５の構成、各分岐条件を示す情報等を学習結果データ１２５として記憶部１２に格納する。 In this embodiment, each leaf node L1 to L5 (class C1 to C5) is associated with a correction value corresponding to the input feature value and attribute value. That is, when the learning processing unit 112 inputs the feature value 1211 and the attribute value 1212, the prediction model 5 is reached so as to reach the leaf node of the class corresponding to the correction value 1213 associated with the input feature value 1211 and attribute value 1212. Build a (decision tree). The learning processing unit 112 stores the configuration of the constructed learned prediction model 5, information indicating each branch condition, and the like in the storage unit 12 as learning result data 125.
　＜制御装置＞
　次に、図８を用いて、本実施形態に係る制御装置２のソフトウェア構成の一例について説明する。図８は、本実施形態に係る制御装置２のソフトウェア構成の一例を模式的に例示する。 <Control device>
Next, an example of the software configuration of the control device 2 according to the present embodiment will be described with reference to FIG. FIG. 8 schematically illustrates an example of the software configuration of the control device 2 according to the present embodiment.
　制御装置２の制御部２１は、記憶部２２に記憶された制御プログラム８２をＲＡＭに展開する。そして、制御部２１は、ＲＡＭに展開された制御プログラム８２をＣＰＵにより解釈及び実行して、制御プログラム８２に含まれる一連の命令に基づいて、各構成要素を制御する。これによって、図８に示されるとおり、本実施形態に係る制御装置２は、入力データ取得部２１１、予測演算部２１２、及び動作制御部２１３をソフトウェアモジュールとして備えるコンピュータとして動作する。すなわち、本実施形態では、各ソフトウェアモジュールは、制御部２１（ＣＰＵ）により実現される。 The control unit 21 of the control device 2 expands the control program 82 stored in the storage unit 22 in the RAM. Then, the control unit 21 interprets and executes the control program 82 expanded in the RAM, and controls each component based on a series of instructions included in the control program 82. Accordingly, as illustrated in FIG. 8, the control device 2 according to the present embodiment operates as a computer including the input data acquisition unit 211, the prediction calculation unit 212, and the operation control unit 213 as software modules. That is, in this embodiment, each software module is realized by the control unit 21 (CPU).
　入力データ取得部２１１は、生産装置３の動作を決定する要因に関する入力データを取得する。本実施形態では、予測モデル５は、ワーク４０の特徴量及び製品４１を生産する環境の属性値の入力に対して、製品４１の生産に適応した指令値を予測するように構築される。そこで、入力データ取得部２１１は、ワーク４０の特徴量７１及び製品４１を生産する環境の属性値７２を入力データとして取得する。 The input data acquisition unit 211 acquires input data relating to factors that determine the operation of the production apparatus 3. In the present embodiment, the prediction model 5 is constructed so as to predict a command value adapted to the production of the product 41 with respect to the input of the feature value of the workpiece 40 and the attribute value of the environment in which the product 41 is produced. Therefore, the input data acquisition unit 211 acquires the feature value 71 of the workpiece 40 and the attribute value 72 of the environment in which the product 41 is produced as input data.
　予測演算部２１２は、学習装置１により生成された学習結果データ１２５を保持している。これにより、予測演算部２１２は、ワーク４０から製品４１を生産する生産装置３への指令値であって、生産装置３による製品４１の生産に適応した指令値を予測するように構築された予測モデル５を備えている。予測演算部２１２は、学習結果データ１２５を参照し、予測制御に利用する予測モデル５の設定を行う。 The prediction calculation unit 212 holds the learning result data 125 generated by the learning device 1. Thereby, the prediction calculation unit 212 is a command value that is a command value from the workpiece 40 to the production apparatus 3 that produces the product 41, and is configured to predict a command value that is adapted to the production of the product 41 by the production apparatus 3. Model 5 is provided. The prediction calculation unit 212 refers to the learning result data 125 and sets the prediction model 5 used for prediction control.
　次に、予測演算部２１２は、取得した入力データ（特徴量７１及び属性値７２）を予測モデル５に入力し、予測モデル５の演算処理を実行する。これにより、予測演算部２１２は、生産装置３による製品４１の生産に適応した指令値を予測した結果に対応する出力値を当該予測モデル５から取得する。予測演算部２１２は、取得した出力値に基づいて、学習装置１によって決定された第２の閾値６２により規定される第２の許容範囲内で、生産装置３に対する指令値を決定する。 Next, the prediction calculation unit 212 inputs the acquired input data (feature value 71 and attribute value 72) to the prediction model 5, and executes the calculation process of the prediction model 5. Thereby, the prediction calculation unit 212 acquires from the prediction model 5 an output value corresponding to a result of predicting a command value adapted to the production of the product 41 by the production apparatus 3. The prediction calculation unit 212 determines a command value for the production apparatus 3 within the second allowable range defined by the second threshold value 62 determined by the learning apparatus 1 based on the acquired output value.
　本実施形態では、予測モデル５は、製品４１の生産に適応した指令値の予測の結果に対応する出力値として、指令値の基準値７０に対する補正値７３を出力する決定木により構成されている。そのため、予測演算部２１２は、予測モデル５の演算処理として、決定木の探索処理を実行する。予測演算部２１２は、この予測モデル５の演算処理を完了することで、当該予測モデル５から補正値７３に対応する出力値を取得することができる。 In the present embodiment, the prediction model 5 is configured by a decision tree that outputs a correction value 73 for the reference value 70 of the command value as an output value corresponding to the result of prediction of the command value adapted to the production of the product 41. . Therefore, the prediction calculation unit 212 executes a decision tree search process as the calculation process of the prediction model 5. The prediction calculation unit 212 can acquire an output value corresponding to the correction value 73 from the prediction model 5 by completing the calculation process of the prediction model 5.
　具体例として、図７Ａに例示される決定木（予測モデル５）の探索処理について説明する。予測演算部２１２は、予測モデル５の根ノードＲから探索処理を開始して、入力データが分岐条件を満たすか否かの判定を繰り返すことで、いずれかの葉ノードＬ１～Ｌ５に到達するまで、より深いノードに探索を進めていく。図７Ａの例では、入力ｘ０が特徴量７１に対応し、入力ｘ１が属性値７２に対応している。図７Ｂは、各入力（ｘ０、ｘ１）と到達する葉ノードＬ１～Ｌ５に対応付けられたクラスＣ１～Ｃ５との関係を例示している。 As a specific example, a search process of the decision tree (prediction model 5) illustrated in FIG. 7A will be described. The prediction calculation unit 212 starts the search process from the root node R of the prediction model 5 and repeats the determination of whether or not the input data satisfies the branch condition until reaching any one of the leaf nodes L1 to L5. , The search proceeds to deeper nodes. In the example of FIG. 7A, the input x0 corresponds to the feature amount 71, and the input x1 corresponds to the attribute value 72. FIG. 7B illustrates the relationship between each input (x0, x1) and the classes C1 to C5 associated with the reaching leaf nodes L1 to L5.
　例えば、入力ｘ０が２０００であり、入力ｘ１が３０であることを想定する。この場合、予測演算部２１２は、予測モデル５の１階層目の演算処理（探索処理）として、根ノードＲに設定された分岐条件を入力ｘ０が満たすか否かを判定する。図７Ａの例では、根ノードＲに設定された分岐条件は「ｘ０＜２５００」であり、入力ｘ０は２０００であるため、予測演算部２１２は、根ノードＲに設定された分岐条件を入力ｘ０は満たすと判定し、次の階層の中間ノードＮ１に探索を進める。 For example, assume that input x0 is 2000 and input x1 is 30. In this case, the prediction calculation unit 212 determines whether or not the input x0 satisfies the branch condition set in the root node R as the calculation processing (search processing) in the first layer of the prediction model 5. In the example of FIG. 7A, since the branch condition set for the root node R is “x0 <2500” and the input x0 is 2000, the prediction calculation unit 212 inputs the branch condition set for the root node R to the input x0. Is satisfied and the search proceeds to the intermediate node N1 of the next hierarchy.
　次に、予測演算部２１２は、予測モデル５の２階層目の演算処理として、中間ノードＮ１に設定された分岐条件を入力ｘ１が満たすか否かを判定する。図７Ａの例では、中間ノードＮ１に設定された分岐条件は「ｘ１＜２０」であり、入力ｘ１が３０であるため、予測演算部２１２は、中間ノードＮ１に設定された分岐条件を入力ｘ１は満たさないと判定し、次の階層の葉ノードＬ２に進む。これにより、決定木の探索処理が葉ノードＬ２に到達するため、予測モデル５の演算処理が完了する。予測演算部２１２は、予測モデル５の演算処理の最終結果として、葉ノードＬ２のクラスＣ２に対応付けられた補正値７３を取得することができる。 Next, the prediction calculation unit 212 determines whether or not the input x1 satisfies the branch condition set in the intermediate node N1 as the calculation processing of the second hierarchy of the prediction model 5. In the example of FIG. 7A, since the branch condition set in the intermediate node N1 is “x1 <20” and the input x1 is 30, the prediction calculation unit 212 inputs the branch condition set in the intermediate node N1 as input x1. Is not satisfied, and the process proceeds to the leaf node L2 of the next hierarchy. Thereby, since the search process of the decision tree reaches the leaf node L2, the calculation process of the prediction model 5 is completed. The prediction calculation unit 212 can acquire the correction value 73 associated with the class C2 of the leaf node L2 as the final result of the calculation process of the prediction model 5.
　各クラスＣ１～Ｃ５に対応付けられた補正値７３を取得する方法は、実施の形態に応じて適宜決定されてよい。例えば、各クラスＣ１～Ｃ５には、補正値が直接的に対応付けられていてもよい。また、例えば、制御装置２は、各クラスＣ１～Ｃ５と補正値との対応関係を示すテーブル形式等の参照情報を記憶部２２に保持していてもよい。この参照情報は、上記予測モデル５の学習過程において生成されてよく、学習結果データ１２５に含まれていてもよい。この場合、予測演算部２１２は、いずれかの葉ノードに到達した後、到達した葉ノードのクラスを参照情報に照合することで、予測モデル５の演算処理の最終結果として、指令値の基準値７０に対する補正値７３を取得することができる。 The method for acquiring the correction value 73 associated with each of the classes C1 to C5 may be appropriately determined according to the embodiment. For example, correction values may be directly associated with the classes C1 to C5. Further, for example, the control device 2 may hold reference information such as a table format indicating the correspondence between the classes C1 to C5 and the correction values in the storage unit 22. This reference information may be generated in the learning process of the prediction model 5 and may be included in the learning result data 125. In this case, after reaching one of the leaf nodes, the prediction calculation unit 212 collates the class of the reached leaf node with the reference information, thereby obtaining the reference value of the command value as the final result of the calculation process of the prediction model 5 A correction value 73 for 70 can be acquired.
　続いて、予測演算部２１２は、取得した補正値７３により基準値７０を補正することで得られた値に基づいて、第２の許容範囲内で指令値７５を決定する。補正値７３により基準値７０を補正することで得られた値が第２の許容範囲内である場合には、予測演算部２１２は、この得られた値を指令値７５に決定する。一方、補正値７３により基準値７０を補正することで得られた値が第２の許容範囲内ではない場合には、予測演算部２１２は、得られた値を修正することで、第２の許容範囲内で指令値７５を決定する。そして、動作制御部２１３は、決定した指令値７５に基づいて、生産装置３の動作を制御する。 Subsequently, the prediction calculation unit 212 determines the command value 75 within the second allowable range based on the value obtained by correcting the reference value 70 with the acquired correction value 73. When the value obtained by correcting the reference value 70 with the correction value 73 is within the second allowable range, the prediction calculation unit 212 determines the obtained value as the command value 75. On the other hand, when the value obtained by correcting the reference value 70 with the correction value 73 is not within the second allowable range, the prediction calculation unit 212 corrects the obtained value to change the second value. The command value 75 is determined within the allowable range. Then, the operation control unit 213 controls the operation of the production apparatus 3 based on the determined command value 75.
　＜その他＞
　学習装置１及び制御装置２の各ソフトウェアモジュールに関しては後述する動作例で詳細に説明する。なお、本実施形態では、学習装置１及び制御装置２の各ソフトウェアモジュールがいずれも汎用のＣＰＵによって実現される例について説明している。しかしながら、以上のソフトウェアモジュールの一部又は全部が、１又は複数の専用のプロセッサにより実現されてもよい。また、学習装置１及び制御装置２それぞれのソフトウェア構成に関して、実施形態に応じて、適宜、ソフトウェアモジュールの省略、置換及び追加が行われてもよい。 <Others>
The software modules of the learning device 1 and the control device 2 will be described in detail in an operation example described later. In the present embodiment, an example is described in which each software module of the learning device 1 and the control device 2 is realized by a general-purpose CPU. However, some or all of the above software modules may be implemented by one or more dedicated processors. Further, regarding the software configurations of the learning device 1 and the control device 2, software modules may be omitted, replaced, and added as appropriate according to the embodiment.
　§３　動作例
　［学習装置］
　次に、図９を用いて、学習装置１の動作例について説明する。図９は、学習装置１の処理手順の一例を例示するフローチャートである。以下で説明する処理手順は、本発明の「学習方法」の一例である。ただし、以下で説明する処理手順は一例に過ぎず、各処理は可能な限り変更されてよい。また、以下で説明する処理手順について、実施の形態に応じて、適宜、ステップの省略、置換、及び追加が可能である。 §3 Example of operation [Learning device]
Next, an operation example of the learning apparatus 1 will be described with reference to FIG. FIG. 9 is a flowchart illustrating an example of a processing procedure of the learning device 1. The processing procedure described below is an example of the “learning method” of the present invention. However, the processing procedure described below is merely an example, and each processing may be changed as much as possible. Further, in the processing procedure described below, steps can be omitted, replaced, and added as appropriate according to the embodiment.
　（ステップＳ１０１）
　ステップＳ１０１では、制御部１１は、学習データ取得部１１１として動作し、予測モデル５の機械学習に利用する複数件の学習用データセット１２１を取得する。各件の学習用データセット１２１は、生産装置３の動作を決定する要因に関する第１データ、及び生産装置３への指令値であって、第１データにより示される要因に適応した指令値に関する第２データの組み合わせにより構成される。 (Step S101)
In step S <b> 101, the control unit 11 operates as the learning data acquisition unit 111 and acquires a plurality of learning data sets 121 used for machine learning of the prediction model 5. Each learning data set 121 includes first data relating to factors that determine the operation of the production apparatus 3 and command values to the production apparatus 3, which are command values adapted to the factors indicated by the first data. It consists of a combination of two data.
　第１データ及び第２データの構成はそれぞれ、対象装置の動作を予測制御するための予測モデル（本実施形態では、予測モデル５）の機械学習に利用可能であれば、実施の形態に応じて適宜決定されてよい。上記のとおり、本実施形態では、第１データは、ワーク４０の特徴量１２１１及び製品４１を生産する環境の属性値１２１２により構成される。また、第２データは、指令値の基準値に対する補正値１２１３であって、特徴量１２１１及び属性値１２１２により示される状況に適応した指令値が得られるように決定された補正値１２１３により構成される。 As long as the configurations of the first data and the second data can be used for machine learning of a prediction model (prediction model 5 in this embodiment) for predictive control of the operation of the target device, each configuration depends on the embodiment. It may be determined as appropriate. As described above, in the present embodiment, the first data includes the feature value 1211 of the workpiece 40 and the attribute value 1212 of the environment in which the product 41 is produced. The second data is a correction value 1213 with respect to the reference value of the command value, and is composed of a correction value 1213 determined so that a command value adapted to the situation indicated by the feature value 1211 and the attribute value 1212 can be obtained. The
　ワーク４０の特徴量１２１１は、ワーク４０の何らかの特徴を示し得るものであれば特に限定されなくてもよく、実施の形態に応じて適宜選択されてよい。また、製品４１を生産する環境の属性値１２１２は、生産装置３が稼動する環境に関する何らかの属性を示し得るものであれば特に限定されなくてもよく、実施の形態に応じて適宜選択されてよい。 The feature amount 1211 of the workpiece 40 may not be particularly limited as long as it can show some feature of the workpiece 40, and may be appropriately selected according to the embodiment. Further, the attribute value 1212 of the environment in which the product 41 is produced may not be particularly limited as long as it can indicate some attribute relating to the environment in which the production apparatus 3 operates, and may be appropriately selected according to the embodiment. .
　本実施形態では、生産装置３は、プレス機である。上記のとおり、生産装置３では、プレス時間が不十分であったり、上側金型３２が下死点に到達するまでサーボモータを駆動していなかったりすると、得られる製品４１の品質が悪化してしまう。そのため、ワーク４０の特徴量１２１１及び製品４１を生産する環境の属性値１２１２はそれぞれ、生産装置３におけるプレス成形の工程に関するものであるのが好ましい。 In the present embodiment, the production apparatus 3 is a press machine. As described above, in the production apparatus 3, if the press time is insufficient or the servo motor is not driven until the upper die 32 reaches the bottom dead center, the quality of the product 41 obtained is deteriorated. End up. For this reason, it is preferable that the feature value 1211 of the workpiece 40 and the attribute value 1212 of the environment for producing the product 41 each relate to a press forming process in the production apparatus 3.
　そこで、ワーク４０の特徴量１２１１には、例えば、硬さ、寸法、材質、重さ、熱等を示すものが選択されてもよい。また、製品４１を生産する環境の属性値１２１２には、例えば、生産装置３の周囲の温度、湿度、装置の劣化度合い（例えば、経年数、加工回数等）、振動等を示すものが選択されてよい。このとき、ワーク４０の特徴量１２１１は、ワーク４０の特徴を直接的に示すものであってもよいし、ワーク４０の特徴を間接的に示すものであってもよい。ワーク４０の特徴を直接的に示すとは、例えば、ワーク４０の硬さ（硬度）そのものを数値、クラス等で表現することである。一方、ワーク４０の特徴を間接的に示すとは、例えば、ワーク４０の硬さ（硬度）を測定する際に得られた２次的指標（例えば、ワークにかかる荷重、測定の際に作用させたトルク等）を数値、クラス等で表現することである。属性値１２１２についても同様である。 Therefore, as the feature quantity 1211 of the workpiece 40, for example, a value indicating hardness, dimensions, material, weight, heat, or the like may be selected. Further, as the attribute value 1212 of the environment in which the product 41 is produced, for example, a value indicating the ambient temperature and humidity of the production apparatus 3, the degree of deterioration of the apparatus (for example, age, the number of times of processing, etc.), vibration, etc. It's okay. At this time, the feature amount 1211 of the workpiece 40 may directly indicate the feature of the workpiece 40 or may indirectly indicate the feature of the workpiece 40. Directly indicating the characteristics of the workpiece 40 is, for example, expressing the hardness (hardness) of the workpiece 40 itself as a numerical value, a class, or the like. On the other hand, indirectly indicating the characteristics of the workpiece 40 is, for example, a secondary index obtained when measuring the hardness (hardness) of the workpiece 40 (for example, acting on the workpiece, the load applied to the workpiece, and the measurement). Torque, etc.) is expressed by numerical value, class, etc. The same applies to the attribute value 1212.
　このような各件の学習用データセット１２１は、実施の形態に応じて適宜生成されてよい。例えば、生産装置３を稼働させて、ワーク４０の特徴量１２１１及び製品４１を生産する環境の属性値１２１２を様々な条件で取得する。特徴量１２１１及び属性値１２１２の取得には、公知のセンサが用いられてよい。一例として、特徴量１２１１としてワーク４０の硬さを取得する場合には、硬度計が用いられてよい。また、属性値１２１２として温度を取得する場合には、温度センサが用いられてよい。そして、得られた特徴量１２１１及び属性値１２１２に対して、この条件において適切な指令値を得るための補正値１２１３を組み合わせる。これにより、各件の学習用データセット１２１を生成することができる。 Such a learning data set 121 for each case may be appropriately generated according to the embodiment. For example, the production apparatus 3 is operated, and the feature value 1211 of the workpiece 40 and the attribute value 1212 of the environment for producing the product 41 are acquired under various conditions. A known sensor may be used to acquire the feature amount 1211 and the attribute value 1212. As an example, when acquiring the hardness of the workpiece 40 as the feature amount 1211, a hardness meter may be used. In addition, when acquiring the temperature as the attribute value 1212, a temperature sensor may be used. Then, the obtained feature value 1211 and attribute value 1212 are combined with a correction value 1213 for obtaining an appropriate command value under this condition. Thereby, each learning data set 121 can be generated.
　この学習用データセット１２１の生成は、学習装置１により行われてもよい。この場合、制御部１１は、オペレータによる入力装置１４の操作に応じて、各件の学習用データセット１２１を生成してもよい。また、制御部１１は、学習プログラム８１の処理により、各件の学習用データセット１２１を自動的に生成してもよい。この生成処理を実行することで、本ステップＳ１０１では、制御部１１は、複数件の学習用データセット１２１を取得することができる。 The generation of the learning data set 121 may be performed by the learning device 1. In this case, the control unit 11 may generate each learning data set 121 in accordance with the operation of the input device 14 by the operator. Further, the control unit 11 may automatically generate the learning data set 121 for each case by the processing of the learning program 81. By executing this generation process, in step S101, the control unit 11 can acquire a plurality of learning data sets 121.
　また、学習用データセット１２１の生成は、学習装置１以外の他の情報処理装置により行われてもよい。他の情報処理装置では、各件の学習用データセット１２１は、オペレータにより手動で生成されてもよいし、プログラムの処理により自動的に生成されてもよい。この場合、本ステップ１０１では、制御部１１は、ネットワーク、記憶媒体９１等を介して、他の情報処理装置により生成された複数件の学習用データセット１２１を取得することができる。 Further, the generation of the learning data set 121 may be performed by an information processing apparatus other than the learning apparatus 1. In another information processing apparatus, the learning data set 121 for each case may be manually generated by an operator or may be automatically generated by processing of a program. In this case, in step 101, the control unit 11 can acquire a plurality of learning data sets 121 generated by other information processing apparatuses via the network, the storage medium 91, and the like.
　なお、本ステップＳ１０１で取得する学習用データセット１２１の件数は、実施の形態に応じて適宜決定されてよく、例えば、決定木の機械学習を実施可能な程度に適宜決定されてよい。これにより、複数件の学習用データセット１２１を取得すると、制御部１１は、次のステップＳ１０２に処理を進める。 Note that the number of learning data sets 121 acquired in step S101 may be determined as appropriate according to the embodiment, and may be determined as appropriate to the extent that machine learning of a decision tree can be performed, for example. Thus, when a plurality of learning data sets 121 are acquired, the control unit 11 advances the processing to the next step S102.
　（ステップＳ１０２）
　ステップＳ１０２では、制御部１１は、学習処理部１１２として動作し、取得した複数件の学習用データセット１２１を利用した機械学習を行うことにより、学習済みの予測モデル５を構築する。 (Step S102)
In step S <b> 102, the control unit 11 operates as the learning processing unit 112 and constructs the learned prediction model 5 by performing machine learning using the plurality of acquired learning data sets 121.
　本実施形態では、制御部１１は、取得した各件の学習用データセット１２１について、特徴量１２１１及び属性値１２１２を入力すると、入力した特徴量１２１１及び属性値１２１２に関連付けられた補正値１２１３に対応する値を出力するように予測モデル５を構築する。より詳細には、制御部１１は、特徴量１２１１及び属性値１２１２に基づいて根ノードから開始して、関連付けられた補正値１２１３に対応するクラスの葉ノードに到達する探索が可能な決定木を構築する。この決定木の学習方法には、ＣＬＳ（Concept Learning System）、ＩＤ３（Iterative Dichotomiser 3）、Ｃ４．５等が用いられてもよい。これにより、制御部１１は、学習済みの予測モデル５を構築することができる。学習済みの予測モデル５を構築すると、制御部１１は、次のステップＳ１０３に処理を進める。 In this embodiment, when the control unit 11 inputs the feature value 1211 and the attribute value 1212 for the acquired learning data set 121, the control unit 11 sets the correction value 1213 associated with the input feature value 1211 and attribute value 1212. The prediction model 5 is constructed so as to output a corresponding value. More specifically, the control unit 11 starts from the root node based on the feature value 1211 and the attribute value 1212 and determines a decision tree that can be searched to reach the leaf node of the class corresponding to the associated correction value 1213. To construct. For this decision tree learning method, CLS (ConceptConLearning System), ID3 (Iterative Dichotomisertom3), C4.5, or the like may be used. Thereby, the control unit 11 can construct the learned prediction model 5. When the learned prediction model 5 is constructed, the control unit 11 advances the processing to the next step S103.
　（ステップＳ１０３及びＳ１０４）
　ステップＳ１０３では、制御部１１は、推定部１１３として動作し、取得した複数件の学習用データセット１２１における第２データの分布６１から指令値の取り得る数値範囲を推定する。ステップＳ１０４では、制御部１１は、閾値決定部１１４として動作し、生産装置３への指令値に対して予め設定された第１の閾値６０により規定される第１の許容範囲を拡げるように、推定した数値範囲に基づいて、生産装置３への指令値に対する第２の閾値６２を決定する。 (Steps S103 and S104)
In step S <b> 103, the control unit 11 operates as the estimation unit 113 and estimates a numerical value range that the command value can take from the distribution 61 of the second data in the acquired plurality of learning data sets 121. In step S104, the control unit 11 operates as the threshold value determination unit 114, and expands the first allowable range defined by the first threshold value 60 set in advance with respect to the command value to the production apparatus 3. Based on the estimated numerical range, the second threshold value 62 for the command value to the production apparatus 3 is determined.
　（Ａ）表現形式
　推定される指令値の取り得る数値範囲、第１の閾値６０、及び第２の閾値６２の表現形式は、特に限定されなくてもよく、実施の形態に応じて適宜決定されてよい。本実施形態では、第２データは、補正値１２１３により構成される。そのため、本ステップＳ１０３では、制御部１１は、補正値１２１３の取り得る数値範囲を推定することにより、指令値のとり得る数値範囲を間接的に推定してもよい。これに応じて、第１の閾値６０及び第２の閾値６２は、補正値に対して設定されることで、指令値の許容範囲を間接的に規定してもよい。また、制御部１１は、補正値１２１３により基準値７０を補正することで得られる値により、指令値の取り得る数値範囲を直接的に推定してもよい。これに応じて、第１の閾値６０及び第２の閾値６２は、指令値に対して設定されることで、当該指令値の許容範囲を直接的に設定されてよい。いずれのケースも同様に取り扱い可能である。以下では、説明の便宜のため、指令値の取り得る数値範囲が直接的に推定されるものとし、第１の閾値６０及び第２の閾値６２は、指令値に対して直接的に設定されるものと想定する。 (A) Expression format The expression range of the estimated command value, the first threshold value 60, and the second threshold value 62 may not be particularly limited, and may be appropriately determined according to the embodiment. It's okay. In the present embodiment, the second data is constituted by a correction value 1213. Therefore, in step S103, the control unit 11 may indirectly estimate the numerical range that the command value can take by estimating the numerical range that the correction value 1213 can take. Accordingly, the first threshold value 60 and the second threshold value 62 may be set for the correction value to indirectly define the allowable range of the command value. Further, the control unit 11 may directly estimate the numerical range that the command value can take, based on a value obtained by correcting the reference value 70 with the correction value 1213. Accordingly, the first threshold 60 and the second threshold 62 may be set for the command value, so that the allowable range of the command value may be set directly. Both cases can be handled in the same way. In the following, for convenience of explanation, it is assumed that a numerical value range that the command value can take is directly estimated, and the first threshold value 60 and the second threshold value 62 are set directly with respect to the command value. Assumes something.
　（Ｂ）数値範囲の推定方法
　次に、ステップＳ１０３において、第２データの分布から指令値の取り得る数値範囲を推定する方法について説明する。制御部１１は、各件の学習用データセット１２１における補正値１２１３（第２データ）を参照することにより、当該補正値１２１３（第２データ）により指定される指令値の分布を把握することができる。このとき、制御部１１は、正規分布、ガンマ分布、指数分布等の統計的手法を用いて、指令値の分布を近似してもよい。 (B) Method of Estimating Numerical Range Next, a method of estimating the numerical range that can be taken by the command value from the distribution of the second data in step S103 will be described. The control unit 11 can grasp the distribution of the command value specified by the correction value 1213 (second data) by referring to the correction value 1213 (second data) in the learning data set 121 of each case. it can. At this time, the control unit 11 may approximate the distribution of the command value using a statistical method such as a normal distribution, a gamma distribution, and an exponential distribution.
　図１０は、正規分布により指令値の分布を近似した場面の一例を示す。正規分布により指令値の分布を近似する方法には、公知の統計処理が用いられてよい。この場合、制御部１１は、近似した正規分布に基づいて、指令値の最小値及び最大値を算出することができる。本ステップＳ１０３では、制御部１１は、この最小値から最大値までの数値範囲を指令値の取り得る数値範囲として推定してもよい。正規分布における指令値の最小値及び最大値はそれぞれ、当該指令値の取り得る数値範囲の境界値の一例である。 FIG. 10 shows an example of a scene in which the distribution of command values is approximated by a normal distribution. A known statistical process may be used as a method of approximating the distribution of command values by a normal distribution. In this case, the control unit 11 can calculate the minimum value and the maximum value of the command value based on the approximated normal distribution. In step S103, the control unit 11 may estimate a numerical range from the minimum value to the maximum value as a numerical range that the command value can take. Each of the minimum value and the maximum value of the command value in the normal distribution is an example of a boundary value in a numerical range that can be taken by the command value.
　なお、指令値の取り得る数値範囲を推定する方法は、このような統計的手法に限定されなくてもよい。上記以外の方法として、例えば、制御部１１は、各件の学習用データセット１２１における補正値１２１３（第２データ）を参照することにより把握した分布をそのまま指令値の取り得る数値範囲として利用してもよい。この場合、学習用データセット１２１の補正値１２１３（第２データ）により指定される指令値の最小値及び最大値がそれぞれ、当該指令値の取り得る数値範囲の境界値となる。制御部１１は、この最小値から最大値までの数値範囲を指令値の取り得る数値範囲として推定してもよい。 Note that the method of estimating the numerical range that the command value can take may not be limited to such a statistical method. As a method other than the above, for example, the control unit 11 uses the distribution obtained by referring to the correction value 1213 (second data) in the learning data set 121 of each case as it is as a numerical range that the command value can take. May be. In this case, the minimum value and the maximum value of the command value specified by the correction value 1213 (second data) of the learning data set 121 are respectively boundary values of numerical values that can be taken by the command value. The control unit 11 may estimate the numerical range from the minimum value to the maximum value as a numerical range that the command value can take.
　（Ｃ）第２の閾値の決定方法
　次に、ステップＳ１０４において、推定した数値範囲に基づいて、第２の閾値６２を決定する方法について説明する。ステップＳ１０３により推定した数値範囲から第２の閾値６２を導出する方法は、実施の形態に応じて適宜設定されてよい。例えば、制御部１１は、当該数値範囲の境界値を利用することで、第２の閾値６２を決定することができる。一例として、制御部１１は、推定した数値範囲の境界値又は第１の閾値６０と当該境界値との間の値を第２の閾値６２として採用してもよい。 (C) Second Threshold Determination Method Next, a method for determining the second threshold 62 based on the estimated numerical range in step S104 will be described. The method for deriving the second threshold value 62 from the numerical range estimated in step S103 may be set as appropriate according to the embodiment. For example, the control unit 11 can determine the second threshold 62 by using the boundary value of the numerical value range. As an example, the control unit 11 may adopt a boundary value in the estimated numerical range or a value between the first threshold value 60 and the boundary value as the second threshold value 62.
　指令値の許容範囲は、下限値及び上限値の少なくとも一方を指定することにより規定することができる。第１の閾値６０は、当該第１の許容範囲の下限値であってもよいし、当該第１の許容範囲の上限値であってもよい。また、第１の許容範囲が下限値及び上限値の両方により規定される場合、当該第１の許容範囲の下限値及び上限値それぞれが第１の閾値６０として取り扱われてもよい。以下の図１１Ａ及び図１１Ｂの例では、説明の便宜のため、第１の許容範囲の下限値及び上限値それぞれが第１の閾値６０として取り扱われるものと想定する。 The allowable range of the command value can be defined by specifying at least one of the lower limit value and the upper limit value. The first threshold 60 may be a lower limit value of the first allowable range, or may be an upper limit value of the first allowable range. When the first allowable range is defined by both the lower limit value and the upper limit value, each of the lower limit value and the upper limit value of the first allowable range may be handled as the first threshold value 60. In the example of FIG. 11A and FIG. 11B below, it is assumed that the lower limit value and the upper limit value of the first allowable range are each handled as the first threshold value 60 for convenience of explanation.
　図１１Ａは、推定した数値範囲の境界値を第２の閾値６２として採用する場面の一例を模式的に例示する。図１１Ｂは、推定した数値範囲の境界値と第１の閾値６０との間の値を第２の閾値６２として採用する場面の一例を模式的に例示する。図１１Ａ及び図１１Ｂの例において、グラフの横軸は、生産装置３への指令値（入力）に対応し、グラフの縦軸は、サーボモータのトルク（出力）に対応する。 FIG. 11A schematically illustrates an example of a scene in which the estimated boundary value of the numerical range is adopted as the second threshold value 62. FIG. 11B schematically illustrates an example of a scene in which a value between the estimated boundary value of the numerical range and the first threshold value 60 is adopted as the second threshold value 62. 11A and 11B, the horizontal axis of the graph corresponds to the command value (input) to the production apparatus 3, and the vertical axis of the graph corresponds to the torque (output) of the servo motor.
　第１の閾値６０は、第２の閾値６２を決定する前に予め与えられる。この第１の閾値６０は、生産装置３を利用するユーザにより予め決定されてもよいし、生産装置３又は制御装置２において予め決定されていてもよい。第１の閾値６０は、予測モデル５により生産装置３を予測制御する場面ではなく、生産装置３又は制御装置２をユーザが手動により操作する場面において、生産装置３の指令値に対する制約条件として利用されてもよい。制御部１１は、ネットワーク等を介して制御装置２又は生産装置３に問い合わせることにより、第１の閾値６０を取得してもよい。また、学習装置１は、第１の閾値６０を記憶部１２等に予め保持していてもよいし、オペレータの指定により第１の閾値６０を取得してもよい。 The first threshold value 60 is given in advance before the second threshold value 62 is determined. The first threshold 60 may be determined in advance by a user who uses the production apparatus 3, or may be determined in advance in the production apparatus 3 or the control apparatus 2. The first threshold value 60 is used as a restriction condition for the command value of the production apparatus 3 in a scene where the user manually operates the production apparatus 3 or the control apparatus 2 instead of the scene where the production apparatus 3 is controlled by the prediction model 5. May be. The control unit 11 may acquire the first threshold 60 by making an inquiry to the control device 2 or the production device 3 via a network or the like. Moreover, the learning apparatus 1 may hold | maintain the 1st threshold value 60 previously in the memory | storage part 12, etc., and may acquire the 1st threshold value 60 by an operator's designation | designated.
　図１１Ａの例では、ステップＳ１０３により推定した数値範囲の最大値が第１の許容範囲の上限値（第１の閾値６０）を超える場合に、本ステップＳ１０４では、制御部１１は、当該最大値を第２の閾値６２として採用する。これにより、制御部１１は、第１の許容範囲の上限値を超える値を第２の閾値６２として採用することができる。この第２の閾値６２は、第２の許容範囲の上限値として取り扱われる。 In the example of FIG. 11A, when the maximum value of the numerical range estimated in step S103 exceeds the upper limit value (first threshold 60) of the first allowable range, in step S104, the control unit 11 determines the maximum value. Is adopted as the second threshold 62. Accordingly, the control unit 11 can employ a value that exceeds the upper limit value of the first allowable range as the second threshold value 62. The second threshold 62 is handled as the upper limit value of the second allowable range.
　また、ステップＳ１０３により推定した数値範囲の最小値が第１の許容範囲の下限値（第１の閾値６０）未満である場合に、本ステップＳ１０４では、制御部１１は、当該最小値を第２の閾値６２として採用する。これにより、制御部１１は、第１の許容範囲の下限値より小さい値を第２の閾値６２として採用することができる。この第２の閾値６２は、第２の許容範囲の下限値として取り扱われる。 Further, when the minimum value of the numerical range estimated in step S103 is less than the lower limit value (first threshold 60) of the first allowable range, in step S104, the control unit 11 sets the minimum value to the second value. The threshold 62 is adopted. Thereby, the control unit 11 can employ a value smaller than the lower limit value of the first allowable range as the second threshold value 62. The second threshold 62 is handled as the lower limit value of the second allowable range.
　一方、図１１Ｂの例では、ステップＳ１０３により推定した数値範囲の最大値が第１の許容範囲の上限値（第１の閾値６０）を超える場合に、本ステップＳ１０４では、制御部１１は、当該最大値と当該上限値との間の値を第２の閾値６２として採用する。このとき、第２の閾値６２として採用する値は、実施の形態に応じて適宜決定されてよい。例えば、制御部１１は、推定した数値範囲の最大値と第１の許容範囲の上限値との平均値を第２の閾値６２として採用してもよい。これにより、制御部１１は、第１の許容範囲の上限値を超える値を第２の閾値６２として採用することができる。この第２の閾値６２は、第２の許容範囲の上限値として取り扱われる。 On the other hand, in the example of FIG. 11B, when the maximum value of the numerical range estimated in step S103 exceeds the upper limit value (first threshold 60) of the first allowable range, in step S104, the control unit 11 A value between the maximum value and the upper limit value is adopted as the second threshold value 62. At this time, the value employed as the second threshold 62 may be appropriately determined according to the embodiment. For example, the control unit 11 may adopt an average value of the estimated maximum value of the numerical range and the upper limit value of the first allowable range as the second threshold value 62. Accordingly, the control unit 11 can employ a value that exceeds the upper limit value of the first allowable range as the second threshold value 62. The second threshold 62 is handled as the upper limit value of the second allowable range.
　また、ステップＳ１０３により推定した数値範囲の最小値が第１の許容範囲の下限値（第１の閾値６０）未満である場合に、本ステップＳ１０４では、制御部１１は、当該最小値と当該下限値との間の値を第２の閾値６２として採用する。このとき、上記上限値のケースと同様に、第２の閾値６２として採用する値は、実施の形態に応じて適宜決定されてよい。例えば、制御部１１は、推定した数値範囲の最小値と第１の許容範囲の下限値との平均値を第２の閾値６２として採用してもよい。これにより、制御部１１は、第１の許容範囲の下限値より小さい値を第２の閾値６２として採用することができる。この第２の閾値６２は、第２の許容範囲の下限値として取り扱われる。 Further, when the minimum value of the numerical range estimated in step S103 is less than the lower limit value (first threshold 60) of the first allowable range, in step S104, the control unit 11 determines that the minimum value and the lower limit A value between the values is adopted as the second threshold value 62. At this time, as in the case of the upper limit value, the value adopted as the second threshold value 62 may be appropriately determined according to the embodiment. For example, the control unit 11 may adopt an average value of the estimated minimum value of the numerical range and the lower limit value of the first allowable range as the second threshold value 62. Thereby, the control unit 11 can employ a value smaller than the lower limit value of the first allowable range as the second threshold value 62. The second threshold 62 is handled as the lower limit value of the second allowable range.
　図１１Ａ及び図１１Ｂに示されるとおり、以上のいずれかの方法で決定された第２の閾値６２により規定される第２の許容範囲は、第１の許容範囲よりも広くなる。したがって、本実施形態に係る制御部１１は、以上のいずれかの方法により、予め設定された第１の閾値６０により規定される第１の許容範囲を拡げるように、ステップＳ１０３により推定された数値範囲に基づいて、指令値に対する第２の閾値６２を決定することができる。 As shown in FIGS. 11A and 11B, the second allowable range defined by the second threshold 62 determined by any of the above methods is wider than the first allowable range. Therefore, the control unit 11 according to the present embodiment uses the numerical values estimated in step S103 so as to widen the first allowable range defined by the preset first threshold 60 by any one of the methods described above. Based on the range, the second threshold 62 for the command value can be determined.
　このとき、制御部１１は、予め設定された安全条件を満たすように第２の閾値６２を決定してもよい。安全条件は、生産装置３の動作を安全に制御可能なように、実施の形態に応じて適宜規定されてよい。制御部１１は、ネットワーク等を介して制御装置２又は生産装置３に問い合わせることにより、この安全条件を示す情報を取得してもよい。また、学習装置１は、安全条件を示す情報を記憶部１２等に予め保持していてもよいし、オペレータの指定により安全条件を示す情報を取得してもよい。 At this time, the control unit 11 may determine the second threshold 62 so as to satisfy a preset safety condition. The safety condition may be appropriately defined according to the embodiment so that the operation of the production apparatus 3 can be safely controlled. The control unit 11 may acquire information indicating the safety condition by making an inquiry to the control device 2 or the production device 3 via a network or the like. Moreover, the learning apparatus 1 may hold | maintain the information which shows safety conditions beforehand in the memory | storage part 12, etc., and may acquire the information which shows safety conditions by an operator's designation | designated.
　例えば、安全条件は、ユーザ、生産装置３の製造者等により予め指定された安全制御用の閾値により規定されてもよい。この安全制御用の閾値が指令値の許容範囲の上限値について設定されている場合、制御部１１は、ステップＳ１０４において決定した値（第２の許容範囲の上限値）が安全制御用の閾値以下であるか否かを判定してもよい。そして、ステップＳ１０４において決定した値が安全制御用の閾値以下であるときには、制御部１１は、当該決定した値を第２の閾値６２（すなわち、第２の許容範囲の上限値）として採用してもよい。一方、そうではないときには、制御部１１は、安全制御用の閾値以下になるように当該値を修正し、修正した値を第２の閾値６２として採用してもよい。 For example, the safety condition may be defined by a threshold value for safety control designated in advance by the user, the manufacturer of the production apparatus 3, or the like. When the threshold value for safety control is set for the upper limit value of the allowable range of the command value, the control unit 11 determines that the value determined in step S104 (the upper limit value of the second allowable range) is equal to or less than the threshold value for safety control. It may be determined whether or not. When the value determined in step S104 is equal to or less than the threshold value for safety control, the control unit 11 adopts the determined value as the second threshold value 62 (that is, the upper limit value of the second allowable range). Also good. On the other hand, when this is not the case, the control unit 11 may correct the value so as to be equal to or less than the threshold value for safety control, and adopt the corrected value as the second threshold value 62.
　同様に、安全制御用の閾値が指令値の許容範囲の下限値について設定されている場合、制御部１１は、ステップＳ１０４において決定した値（第２の許容範囲の下限値）が安全制御用の閾値以上であるか否かを判定してもよい。そして、ステップＳ１０４において決定した値が安全制御用の閾値以上であるときには、制御部１１は、当該決定した値を第２の閾値６２（すなわち、第２の許容範囲の下限値）として採用してもよい。一方、そうではないときには、制御部１１は、安全制御用の閾値以上になるように当該値を修正し、修正した値を第２の閾値６２として採用してもよい。 Similarly, when the threshold value for safety control is set for the lower limit value of the allowable range of the command value, the control unit 11 determines that the value determined in step S104 (the lower limit value of the second allowable range) is for safety control. You may determine whether it is more than a threshold value. When the value determined in step S104 is equal to or greater than the threshold value for safety control, the control unit 11 adopts the determined value as the second threshold value 62 (that is, the lower limit value of the second allowable range). Also good. On the other hand, when this is not the case, the control unit 11 may correct the value so as to be equal to or higher than the threshold value for safety control, and adopt the corrected value as the second threshold value 62.
　また、例えば、安全条件は、生産装置３の動作をシミュレーションする又は生産装置３を実際に駆動することにより規定されてもよい。この場合、制御部１１は、シミュレーション又は実際の駆動の結果に基づいて、ステップＳ１０４において第２の閾値６２として決定した値を指令値として採用した場合に、生産装置３を安全に動作させることができるか否かを判定してもよい。そして、生産装置３を安全に動作させることができると判定した場合、制御部１１は、ステップＳ１０４において決定した値を第２の閾値６２として採用してもよい。一方、生産装置３を安全に動作させることができないと判定した場合には、制御部１１は、生産装置３を安全に動作可能なように当該値を修正し、修正した値を第２の閾値６２として採用してもよい。第２の閾値６２を決定すると、制御部１１は、次のステップＳ１０５に処理を進める。 Also, for example, the safety condition may be defined by simulating the operation of the production apparatus 3 or actually driving the production apparatus 3. In this case, the control unit 11 can safely operate the production apparatus 3 when the value determined as the second threshold value 62 in step S104 is adopted as the command value based on the result of simulation or actual driving. It may be determined whether or not it is possible. And when it determines with the production apparatus 3 being able to operate | move safely, the control part 11 may employ | adopt the value determined in step S104 as the 2nd threshold value 62. FIG. On the other hand, if it is determined that the production apparatus 3 cannot be operated safely, the control unit 11 corrects the value so that the production apparatus 3 can be operated safely, and sets the corrected value to the second threshold value. You may employ | adopt as 62. When the second threshold value 62 is determined, the control unit 11 advances the processing to the next step S105.
　なお、ステップＳ１０３により推定した数値範囲の最大値が第１の許容範囲の上限値以下である場合には、制御部１１は、上記いずれかの方法による第２の許容範囲の上限値（第２の閾値６２）を決定する処理を省略してもよい。同様に、ステップＳ１０３により推定した数値範囲の最小値が第１の許容範囲の下限値以上である場合には、制御部１１は、上記いずれかの方法による第２の閾値６２を決定する処理を省略してもよい。また、第２の許容範囲の上限値及び下限値を決定する方法は互いに異なっていてもよい。例えば、制御部１１は、第２の許容範囲の上限値を決定する方法に図１１Ａにより示される方法を採用し、第２の許容範囲の下限値を決定する方法に図１１Ｂにより示される方法を採用してもよい。 When the maximum value of the numerical range estimated in step S103 is less than or equal to the upper limit value of the first allowable range, the control unit 11 determines the upper limit value of the second allowable range (second The process of determining the threshold value 62) may be omitted. Similarly, when the minimum value of the numerical value range estimated in step S103 is equal to or greater than the lower limit value of the first allowable range, the control unit 11 performs a process of determining the second threshold value 62 by any one of the above methods. It may be omitted. Moreover, the method of determining the upper limit value and the lower limit value of the second allowable range may be different from each other. For example, the control unit 11 adopts the method shown in FIG. 11A as the method for determining the upper limit value of the second allowable range, and uses the method shown in FIG. 11B as the method of determining the lower limit value of the second allowable range. It may be adopted.
　また、図１１Ａ及び図１１Ｂは、生産装置３への指令値（入力）と上側金型３２を駆動するサーボモータのトルク（出力）との関係の一例を模式的に例示している。上記では、第１の閾値６０及び第２の閾値６２は、指令値に対して設定されている。しかしながら、第１の閾値６０及び第２の閾値６２の形式は、このような例に限定されなくてもよい。例えば、第１の閾値６０及び第２の閾値６２はそれぞれ、対象装置の出力（本実施形態では、サーボモータのトルク）に対して設定されることで、指令値の許容範囲を間接的に規定してもよい。 11A and 11B schematically illustrate an example of the relationship between the command value (input) to the production apparatus 3 and the torque (output) of the servo motor that drives the upper mold 32. In the above, the first threshold value 60 and the second threshold value 62 are set for the command value. However, the format of the first threshold value 60 and the second threshold value 62 may not be limited to such an example. For example, each of the first threshold value 60 and the second threshold value 62 is set for the output of the target device (in this embodiment, the torque of the servo motor), thereby indirectly specifying the allowable range of the command value. May be.
　（ステップＳ１０５）
　ステップＳ１０５では、制御部１１は、学習処理部１１２として動作し、機械学習により構築した決定木（学習済みの予測モデル５）の構成及び各分岐条件を示す情報を学習結果データ１２５として記憶部１２に格納する。また、制御部１１は、閾値決定部１１４として動作し、ステップＳ１０４で決定した第２の閾値６２を記憶部１２に格納する。これにより、制御部１１は、本動作例に係る学習処理を終了する。 (Step S105)
In step S105, the control unit 11 operates as the learning processing unit 112, and stores the information indicating the configuration of the decision tree (learned prediction model 5) constructed by machine learning and each branch condition as learning result data 125. To store. The control unit 11 operates as the threshold value determination unit 114 and stores the second threshold value 62 determined in step S104 in the storage unit 12. Thereby, the control part 11 complete | finishes the learning process which concerns on this operation example.
　なお、制御部１１は、上記ステップＳ１０５の処理が完了した後、生成した学習結果データ１２５及び第２の閾値６２を制御装置２に転送してもよい。また、制御部１１は、上記ステップＳ１０１～Ｓ１０５の学習処理を定期的に実行することで、学習結果データ１２５及び第２の閾値６２を定期的に更新してもよい。そして、制御部１１は、生成した学習結果データ１２５及び第２の閾値６２を学習処理の実行毎に制御装置２に転送することで、制御装置２の保持する学習結果データ１２５及び第２の閾値６２を定期的に更新してもよい。また、例えば、制御部１１は、生成した学習結果データ１２５及び第２の閾値６２をＮＡＳ（Network Attached Storage）等のデータサーバに保管してもよい。この場合、制御装置２は、このデータサーバから学習結果データ１２５及び第２の閾値６２を取得してもよい。また、学習装置１により生成された学習結果データ１２５及び第２の閾値６２は、制御装置２に予め組み込まれてもよい。 The control unit 11 may transfer the generated learning result data 125 and the second threshold value 62 to the control device 2 after the process of step S105 is completed. Further, the control unit 11 may periodically update the learning result data 125 and the second threshold value 62 by periodically executing the learning process of steps S101 to S105. And the control part 11 transfers the learning result data 125 and the 2nd threshold value 62 which were produced | generated to the control apparatus 2 for every execution of a learning process, The learning result data 125 and 2nd threshold value which the control apparatus 2 hold | maintains 62 may be updated periodically. For example, the control unit 11 may store the generated learning result data 125 and the second threshold 62 in a data server such as NAS (Network Attached Storage). In this case, the control device 2 may acquire the learning result data 125 and the second threshold value 62 from this data server. Further, the learning result data 125 and the second threshold value 62 generated by the learning device 1 may be incorporated in the control device 2 in advance.
　［制御装置］
　次に、図１２を用いて、運用のフェーズにおける制御装置２の動作例について説明する。図１２は、制御装置２の処理手順の一例を示すフローチャートである。なお、以下で説明する処理手順は一例に過ぎず、各処理は可能な限り変更されてよい。また、以下で説明する処理手順について、実施の形態に応じて、適宜、ステップの省略、置換、及び追加が可能である。 [Control device]
Next, an operation example of the control device 2 in the operation phase will be described with reference to FIG. FIG. 12 is a flowchart illustrating an example of a processing procedure of the control device 2. Note that the processing procedure described below is merely an example, and each processing may be changed as much as possible. Further, in the processing procedure described below, steps can be omitted, replaced, and added as appropriate according to the embodiment.
　（ステップＳ２０１）
　ステップＳ２０１では、制御部２１は、入力データ取得部２１１として動作し、運用フェーズにおいて、要因に関する入力データを取得する。 (Step S201)
In step S201, the control unit 21 operates as the input data acquisition unit 211, and acquires input data regarding factors in the operation phase.
　本実施形態では、上記のとおり、予測モデル５は、ワーク４０の特徴量１２１１及び製品４１を生産する環境の属性値１２１２を第１データとして含む学習用データセット１２１を利用した機械学習により構築される。そのため、本ステップＳ２０１では、制御部２１は、ワーク４０の特徴量７１及び製品４１を生産する環境の属性値７２を取得する。 In the present embodiment, as described above, the prediction model 5 is constructed by machine learning using the learning data set 121 including the feature value 1211 of the workpiece 40 and the attribute value 1212 of the environment in which the product 41 is produced as the first data. The Therefore, in step S201, the control unit 21 acquires the feature value 71 of the workpiece 40 and the attribute value 72 of the environment in which the product 41 is produced.
　特徴量７１及び属性値７２は、上記特徴量１２１１及び属性値１２１２と同種であればよい。また、特徴量７１及び属性値７２それぞれを取得する方法は、実施の形態に応じて適宜選択されてよい。例えば、ワーク４０の特徴量７１（例えば、硬さ等）及び環境の属性値７２（例えば、温度等）それぞれを測定可能に構成された各種センサが生産装置３に配置されてもよい。各種センサには、測定対象となる特徴量７１及び属性値７２の種類に応じて公知のセンサが適宜用いられてよい。この場合、制御部２１は、生産装置３に配置された各種センサから、特徴量７１及び属性値７２それぞれを取得することができる。特徴量７１及び属性値７２それぞれを取得すると、制御部２１は、次のステップＳ２０２に処理を進める。 The feature amount 71 and the attribute value 72 may be the same type as the feature amount 1211 and the attribute value 1212. Further, the method for acquiring the feature amount 71 and the attribute value 72 may be appropriately selected according to the embodiment. For example, various sensors configured to be able to measure the feature amount 71 (for example, hardness) of the workpiece 40 and the environmental attribute value 72 (for example, temperature) may be arranged in the production apparatus 3. As the various sensors, known sensors may be appropriately used according to the types of the feature amount 71 and the attribute value 72 to be measured. In this case, the control unit 21 can acquire each of the feature value 71 and the attribute value 72 from various sensors arranged in the production apparatus 3. When the feature amount 71 and the attribute value 72 are acquired, the control unit 21 proceeds to the next step S202.
　（ステップＳ２０２）
　ステップＳ２０２では、制御部２１は、予測演算部２１２として動作し、取得した入力データ（特徴量７１及び属性値７２）を予測モデル５に入力し、当該予測モデル５の演算処理を実行する。これにより、制御部２１は、生産装置３による製品４１の生産に適応した指令値を予測した結果に対応する出力値を当該予測モデル５から取得する。 (Step S202)
In step S202, the control unit 21 operates as the prediction calculation unit 212, inputs the acquired input data (feature value 71 and attribute value 72) to the prediction model 5, and executes the calculation process of the prediction model 5. Thereby, the control unit 21 acquires an output value corresponding to a result of predicting a command value adapted to the production of the product 41 by the production apparatus 3 from the prediction model 5.
　本実施形態では、予測モデル５は決定木により構成されており、予測モデル５の構成及び各経路の分岐条件を示す情報は、学習結果データ１２５に含まれている。そこで、制御部２１は、学習結果データ１２５を参照することで、予測モデル５の設定を行う。この設定処理によって、制御部２１は、決定木（予測モデル５）の探索処理を開始することができる状態になる。 In the present embodiment, the prediction model 5 is configured by a decision tree, and information indicating the configuration of the prediction model 5 and the branch condition of each path is included in the learning result data 125. Therefore, the control unit 21 sets the prediction model 5 by referring to the learning result data 125. By this setting process, the control unit 21 is in a state where the search process for the decision tree (prediction model 5) can be started.
　次に、制御部２１は、決定木（予測モデル５）の根ノードから葉ノードに向けてリンクをたどる探索処理を実行する。具体的には、探索処理を１度も実行していない場合、制御部２１は、決定木の探索処理として、根ノードに設定された分岐条件を入力データ（特徴量７１及び属性値７２）が満たすか否かを判定する。そして、この判定結果に基づいて、制御部２１は、２段階目の該当ノード（図７Ａの例では、中間ノードＮ１又は中間ノードＮ２）に探索を進める。 Next, the control unit 21 executes a search process that follows the link from the root node of the decision tree (prediction model 5) toward the leaf node. Specifically, when the search process has never been executed, the control unit 21 uses the branch condition set in the root node as input data (feature value 71 and attribute value 72) as the search process of the decision tree. It is determined whether or not it is satisfied. Based on the determination result, the control unit 21 advances the search to the corresponding node in the second stage (in the example of FIG. 7A, the intermediate node N1 or the intermediate node N2).
　同様に、探索処理をｎ回実行した場合（ｎは、１以上の自然数）、探索は、ｎ＋１段目の中間ノードまで探索が進行している。この場合には、制御部２１は、ｎ＋１段目の該当の中間ノードに設定された分岐条件を入力データが満たすか否かを判定する。そして、この判定結果に基づいて、制御部２１は、ｎ＋２段目の該当ノードに探索を進める。 Similarly, when the search process is executed n times (n is a natural number of 1 or more), the search is progressing to the intermediate node at the (n + 1) th stage. In this case, the control unit 21 determines whether or not the input data satisfies the branch condition set in the corresponding intermediate node in the (n + 1) th stage. Based on the determination result, the control unit 21 advances the search to the corresponding node in the (n + 2) th stage.
　決定木のいずれかの葉ノードまでこの探索処理が到達することで、予測モデル５の演算処理が完了する。本実施形態では、予測モデル５を構成する決定木の各葉ノードには、製品４１の生産に適応した指令値の予測の結果に対応する出力値として、指令値の基準値７０に対する補正値７３が対応付けられている。そのため、この予測モデル５の演算処理が完了することで、制御部２１は、予測モデルからの出力値として、探索処理の到達した葉ノードに対応付けられた補正値７３を取得することができる。この出力値を取得すると、制御部２１は、次のステップＳ２０３に処理を進める。 When the search process reaches any leaf node in the decision tree, the calculation process of the prediction model 5 is completed. In this embodiment, a correction value 73 for the reference value 70 of the command value is output to each leaf node of the decision tree constituting the prediction model 5 as an output value corresponding to the prediction result of the command value adapted to the production of the product 41. Are associated. Therefore, when the calculation process of the prediction model 5 is completed, the control unit 21 can acquire the correction value 73 associated with the leaf node to which the search process has reached as the output value from the prediction model. When the output value is acquired, the control unit 21 proceeds to the next step S203.
　（ステップＳ２０３）
　ステップＳ２０３では、制御部２１は、予測演算部２１２として動作し、予測モデル５から取得した出力値に基づいて、学習装置１により決定された第２の閾値６２により規定される第２の許容範囲内で、生産装置３への指令値を決定する。 (Step S203)
In step S <b> 203, the control unit 21 operates as the prediction calculation unit 212 and is based on the output value acquired from the prediction model 5, and the second allowable range defined by the second threshold 62 determined by the learning device 1. The command value to the production apparatus 3 is determined.
　本実施形態では、上記ステップＳ２０２において、制御部２１は、予測モデル５からの出力値として、基準値７０に対する補正値７３を取得している。そのため、制御部２１は、取得した補正値７３で基準値７０を補正（例えば、加算、減算）することで、指令値の予測値を算出する。そして、制御部２１は、算出した予測値が、第２の閾値６２により規定される第２の許容範囲内であるか否かを判定する。 In the present embodiment, in step S202, the control unit 21 acquires a correction value 73 for the reference value 70 as an output value from the prediction model 5. Therefore, the control unit 21 calculates the predicted value of the command value by correcting (for example, adding or subtracting) the reference value 70 with the acquired correction value 73. Then, the control unit 21 determines whether the calculated predicted value is within the second allowable range defined by the second threshold 62.
　算出した予測値が第２の許容範囲内である場合、制御部２１は、算出した予測値を指令値７５に決定する。一方、算出した予測値が第２の許容範囲内ではない場合、制御部２１は、第２の許容範囲内になるように、算出した予測値を適宜修正し、修正した値を指令値７５に決定する。例えば、算出した予測値が第２の許容範囲の上限値を超えている場合、制御部２１は、第２の許容範囲の上限値を指令値７５に決定してもよい。また、例えば、算出した予測値が第２の許容範囲の下限値よりも小さい場合、制御部２１は、第２の許容範囲の下限値を指令値７５に決定してもよい。これにより、第２の許容範囲内で指令値７５を決定すると、制御部２１は、次のステップＳ２０４に処理を進める。 When the calculated predicted value is within the second allowable range, the control unit 21 determines the calculated predicted value as the command value 75. On the other hand, when the calculated predicted value is not within the second allowable range, the control unit 21 appropriately corrects the calculated predicted value so as to be within the second allowable range, and sets the corrected value to the command value 75. decide. For example, when the calculated predicted value exceeds the upper limit value of the second allowable range, the control unit 21 may determine the command value 75 as the upper limit value of the second allowable range. For example, when the calculated predicted value is smaller than the lower limit value of the second allowable range, the control unit 21 may determine the lower limit value of the second allowable range as the command value 75. Accordingly, when the command value 75 is determined within the second allowable range, the control unit 21 proceeds to the next step S204.
　（ステップＳ２０４）
　ステップＳ２０４では、制御部２１は、動作制御部２１３として動作し、決定した指令値７５に基づいて、生産装置３の動作を制御する。指令値７５に基づいて生産装置３の動作を制御する方法は、指令値の形式に応じて適宜選択されてよい。 (Step S204)
In step S <b> 204, the control unit 21 operates as the operation control unit 213 and controls the operation of the production apparatus 3 based on the determined command value 75. A method for controlling the operation of the production apparatus 3 based on the command value 75 may be appropriately selected according to the format of the command value.
　本実施形態では、生産装置３は、プレス機であり、上側金型３２を駆動するサーボドライバ３１を備えている。そのため、指令値７５は、サーボモータの駆動量を規定したパルス数を示してもよい。この場合、制御部２１は、外部インタフェース２４を介して、生産装置３のサーボドライバ３１に対して指令値７５を送信する。サーボドライバ３１は、制御装置２から受信した指令値７５に基づいてサーボモータを駆動する。これにより、制御部２１は、決定した指令値７５に基づいて、生産装置３の動作を制御することができる。生産装置３の動作を制御すると、制御部２１は、本動作例に係る処理を終了する。 In this embodiment, the production apparatus 3 is a press machine and includes a servo driver 31 that drives the upper mold 32. Therefore, the command value 75 may indicate the number of pulses that defines the drive amount of the servo motor. In this case, the control unit 21 transmits a command value 75 to the servo driver 31 of the production apparatus 3 via the external interface 24. The servo driver 31 drives the servo motor based on the command value 75 received from the control device 2. Thereby, the control part 21 can control operation | movement of the production apparatus 3 based on the determined command value 75. FIG. When the operation of the production apparatus 3 is controlled, the control unit 21 ends the process according to this operation example.
　なお、指令値７５の形式は、このような例に限定されなくてもよい。指令値７５は、例えば、サーボモータの駆動量、上側金型３２の移動量等の中間指標により表現されてもよい。この場合、制御部２１は、中間指標により表現された指令値７５をそのまま生産装置３に送信してもよいし、中間指標により表現された指令値７５を、パルス数等の直接利用可能な形式に変換し、変換した指令値７５を生産装置３に送信してもよい。 In addition, the format of the command value 75 may not be limited to such an example. The command value 75 may be expressed by an intermediate index such as a drive amount of the servo motor and a movement amount of the upper mold 32, for example. In this case, the control unit 21 may transmit the command value 75 expressed by the intermediate index as it is to the production apparatus 3, or the command value 75 expressed by the intermediate index can be used directly such as the number of pulses. The converted command value 75 may be transmitted to the production apparatus 3.
　（終了後）
　以上により、制御部２１は、本動作例に係る生産装置３の動作を制御する一連の処理を終了する。制御部２１は、この一連の処理を繰り返し実行することで、生産装置３の動作を継続的に制御することができる。 (After the end)
Thus, the control unit 21 ends the series of processes for controlling the operation of the production apparatus 3 according to this operation example. The control unit 21 can continuously control the operation of the production apparatus 3 by repeatedly executing this series of processes.
　なお、制御装置２は、上記予測モデル５を利用して生産装置３の動作を予測制御するモード（予測制御モード）と、ユーザの操作に応じて生産装置３の動作を制御するモード（手動制御モード）とを切り替え可能に構成されてよい。この場合、動作モードが予測制御モードに設定された場合に、制御部２１は、上記ステップＳ２０１～Ｓ２０４の一連の処理を実行してもよい。また、動作モードが手動制御モードに設定された場合には、制御部２１は、ユーザから指令値の指定を受け付けて、指定された指令値に基づいて、生産装置３の動作を制御してもよい。このとき、制御部２１は、手動制御モードでは、第１の閾値６０を制約条件として利用してもよい。つまり、制御部２１は、第１の閾値６０により規定される第１の許容範囲を超える値の指定を受け付けず、第１の許容範囲内で指令値の指定を受け付けてもよい。 The control device 2 uses the prediction model 5 for predictive control of the operation of the production device 3 (predictive control mode) and a mode for controlling the operation of the production device 3 in accordance with a user operation (manual control). Mode). In this case, when the operation mode is set to the predictive control mode, the control unit 21 may execute a series of processes in steps S201 to S204. In addition, when the operation mode is set to the manual control mode, the control unit 21 receives the designation of the command value from the user, and controls the operation of the production apparatus 3 based on the designated command value. Good. At this time, the control unit 21 may use the first threshold 60 as a constraint condition in the manual control mode. That is, the control unit 21 may accept designation of a command value within the first allowable range without accepting designation of a value exceeding the first allowable range defined by the first threshold 60.
　［特徴］
　以上のとおり、本実施形態では、上記ステップＳ２０３において、生産装置３への指令値を決定する際に、予め設定された第１の閾値６０により規定される第１の許容範囲ではなく、上記ステップＳ１０４により第１の許容範囲を拡げるように設定された第２の閾値６２により規定される第２の許容範囲が指令値の制約条件として利用される。これにより、安全性を過度に考慮して第１の許容範囲が狭く設定されていた場合であっても、生産装置３の動作の制御に用いる指令値７５を許容する範囲を拡げることができる。つまり、上記ステップＳ２０４において、第１の許容範囲を制約条件として利用した場合には拒絶されるような指令値７５の一部を生産装置３の動作の制御に用いるようにすることができる。 [Characteristic]
As described above, in the present embodiment, when determining the command value to the production apparatus 3 in the above step S203, it is not the first allowable range defined by the preset first threshold 60 but the above step. The second permissible range defined by the second threshold value 62 set so as to expand the first permissible range in S104 is used as a constraint condition for the command value. Thereby, even if it is a case where the 1st permissible range was set narrowly in consideration of safety too much, the range which accepts command value 75 used for control of operation of production device 3 can be expanded. That is, in step S204, a part of the command value 75 that is rejected when the first allowable range is used as a constraint condition can be used for controlling the operation of the production apparatus 3.
　更に、上記ステップＳ１０１では、各件の学習用データセット１２１は、特定のケースに適した動作の制御を実現するように収集される。そのため、各件の学習用データセット１２１の第２データ（補正値１２１３）により指定される指令値によれば、生産装置３の動作を安全に制御することができる。よって、上記ステップＳ１０４では、学習用データセット１２１における第２データの分布から推定される数値範囲に基づくことで、生産装置３の動作の安全性を確保するように、第２の許容範囲を規定する第２の閾値６２を決定することができる。特に、上記ステップＳ１０４において、予め設定された安全条件を満たすように第２の閾値６２を決定するようにすることで、生産装置３の動作の安全性を確実に確保することができる。したがって、本実施形態に係る制御システム１００によれば、生産装置３の動作の安全性を確保しつつ、予測モデル５の性能を十分に発揮可能な予測制御を実施することができる。 Furthermore, in the above step S101, the learning data set 121 for each case is collected so as to realize operation control suitable for a specific case. Therefore, according to the command value specified by the second data (correction value 1213) of the learning data set 121 for each case, the operation of the production apparatus 3 can be controlled safely. Therefore, in step S104, the second allowable range is defined so as to ensure the safety of the operation of the production apparatus 3 based on the numerical range estimated from the distribution of the second data in the learning data set 121. A second threshold 62 can be determined. In particular, the safety of the operation of the production apparatus 3 can be reliably ensured by determining the second threshold value 62 so as to satisfy a preset safety condition in step S104. Therefore, according to the control system 100 according to the present embodiment, it is possible to perform predictive control that can sufficiently exhibit the performance of the predictive model 5 while ensuring the safety of the operation of the production apparatus 3.
　§４　変形例
　以上、本発明の実施の形態を詳細に説明してきたが、前述までの説明はあらゆる点において本発明の例示に過ぎない。本発明の範囲を逸脱することなく種々の改良や変形を行うことができることは言うまでもない。例えば、以下のような変更が可能である。なお、以下では、上記実施形態と同様の構成要素に関しては同様の符号を用い、上記実施形態と同様の点については、適宜説明を省略した。以下の変形例は適宜組み合わせ可能である。 §4 Modifications Embodiments of the present invention have been described in detail above, but the above description is merely an illustration of the present invention in all respects. It goes without saying that various improvements and modifications can be made without departing from the scope of the present invention. For example, the following changes are possible. In the following, the same reference numerals are used for the same components as in the above embodiment, and the description of the same points as in the above embodiment is omitted as appropriate. The following modifications can be combined as appropriate.
　＜４．１＞
　上記実施形態では、第１データは、ワーク４０の特徴量１２１１及び製品４１を生産する環境の属性値１２１２の両方により構成されている。これに応じて、予測モデル５の入力には、ワーク４０の特徴量７１及び製品４１を生産する環境の属性値７２の両方が利用されている。しかしながら、予測モデル５の入力は、このような例に限定されなくてもよい。 <4.1>
In the above embodiment, the first data is composed of both the feature value 1211 of the workpiece 40 and the attribute value 1212 of the environment in which the product 41 is produced. Accordingly, both the feature value 71 of the work 40 and the attribute value 72 of the environment in which the product 41 is produced are used for the input of the prediction model 5. However, the input of the prediction model 5 may not be limited to such an example.
　例えば、ワーク４０の特徴量及び製品を生産する環境の属性値のうちの一方は省略されてよい。すなわち、上記第１データは、ワーク４０の特徴量１２１１及び製品４１を生産する環境の属性値１２１２の少なくとも一方により構成されてよい。これに応じて、予測モデル５は、ワーク４０の特徴量７１及び製品４１を生産する環境の属性値７２の少なくとも一方の入力に対して、製品４１の生産に適応した指令値を予測するように構築されてよい。 For example, one of the feature value of the workpiece 40 and the attribute value of the environment in which the product is produced may be omitted. That is, the first data may be configured by at least one of the feature value 1211 of the workpiece 40 and the attribute value 1212 of the environment in which the product 41 is produced. In response to this, the prediction model 5 predicts a command value adapted to the production of the product 41 with respect to at least one input of the feature value 71 of the workpiece 40 and the attribute value 72 of the environment in which the product 41 is produced. May be built.
　また、第１データは、対象装置の動作を決定し得るあらゆる種類の要因に関するデータであってよい。制御装置２は、上記生産装置３以外の種類の対象装置を制御するように構成されてよい。これらに応じて、予測モデル５は、第１データと同じ種類のデータの入力に対して、その入力データにより示される状況に適応した指令値を予測するように構築されてよい。 Further, the first data may be data relating to all types of factors that can determine the operation of the target device. The control device 2 may be configured to control a target device of a type other than the production device 3. Accordingly, the prediction model 5 may be constructed to predict a command value adapted to the situation indicated by the input data with respect to the input of the same type of data as the first data.
　＜４．２＞
　上記実施形態では、第２データは、指令値の基準値に対する補正値１２１３により構成されており、これに応じて、予測モデル５は、指令値の基準値７０に対する補正値７３を出力するように構成されている。しかしながら、予測モデル５の出力形式は、このような例に限定されなくてもよく、実施の形態に応じて適宜決定されてよい。例えば、予測モデル５は、指令値そのものを出力するように構成されてもよい。この場合、第２データは、指令値そのものにより構成されてよい。 <4.2>
In the said embodiment, 2nd data is comprised by the correction value 1213 with respect to the reference value of command value, and according to this, the prediction model 5 outputs the correction value 73 with respect to the reference value 70 of command value. It is configured. However, the output format of the prediction model 5 may not be limited to such an example, and may be appropriately determined according to the embodiment. For example, the prediction model 5 may be configured to output the command value itself. In this case, the second data may be configured by the command value itself.
　＜４．３＞
　上記実施形態では、予測モデル５は、決定木により構成されている。しかしながら、予測モデル５の構成は、予測処理を実行する時点よりも先の時点（将来の時点）における対象装置（一例では、上記生産装置３）への指令値を予測可能であれば、このような例に限定されなくてもよく、実施の形態に応じて適宜選択されてよい。予測モデル５は、例えば、ニューラルネットワーク、サポートベクタマシン等の決定木以外の学習モデルが用いられてもよい。また、予測モデル５には、学習モデル以外のモデル（例えば、所定の関数）が用いられてもよい。 <4.3>
In the above embodiment, the prediction model 5 is configured by a decision tree. However, if the configuration of the prediction model 5 can predict a command value to the target device (in the example, the production device 3 in the example) at a time point (future time point) before the time point at which the prediction process is executed, this is the case. The present invention is not limited to such an example, and may be appropriately selected according to the embodiment. As the prediction model 5, for example, a learning model other than a decision tree such as a neural network or a support vector machine may be used. Further, a model other than the learning model (for example, a predetermined function) may be used as the prediction model 5.
　１…学習装置、
　１１…制御部、１２…記憶部、１３…通信インタフェース、
　１４…入力装置、１５…出力装置、１６…ドライブ、
　１１１…学習データ取得部、１１２…学習処理部、
　１１３…推定部、１１４…閾値決定部、
　８１…学習プログラム、１２１…学習用データセット、
　１２５…学習結果データ、
　９１…記憶媒体、
　２…制御装置、
　２１…制御部、２２…記憶部、２３…通信インタフェース、
　２４…外部インタフェース、
　２５…入力装置、２６…出力装置、２７…ドライブ、
　２１１…入力データ取得部、２１２…予測演算部、
　２１３…動作制御部、
　８２…制御プログラム、９２…記憶媒体、
　３…生産装置（対象装置）、
　３１…サーボドライバ、３２…上金型、３３…下金型、
　４０…ワーク、４１…製品、
　５…予測モデル（決定木）、
　６０…第１の閾値、６１…分布、６２…第２の閾値、
　７０…（指令値の）基準値、
　７１…特徴量、７２…属性値、７３…補正値 1 ... Learning device,
11 ... Control unit, 12 ... Storage unit, 13 ... Communication interface,
14 ... input device, 15 ... output device, 16 ... drive,
111 ... Learning data acquisition unit, 112 ... Learning processing unit,
113 ... Estimating unit, 114 ... Threshold determining unit,
81 ... learning program, 121 ... learning data set,
125 ... learning result data,
91 ... Storage medium,
2 ... Control device,
21 ... Control unit, 22 ... Storage unit, 23 ... Communication interface,
24 ... External interface,
25 ... Input device, 26 ... Output device, 27 ... Drive,
211 ... Input data acquisition unit, 212 ... Prediction calculation unit,
213 ... Operation control unit,
82 ... Control program, 92 ... Storage medium,
3 ... Production equipment (target equipment),
31 ... Servo driver, 32 ... Upper die, 33 ... Lower die,
40 ... work, 41 ... product,
5 ... Prediction model (decision tree),
60 ... first threshold, 61 ... distribution, 62 ... second threshold,
70 ... Reference value (of command value),
71 ... feature amount, 72 ... attribute value, 73 ... correction value

Claims (12)

  1. 　対象装置の動作を決定する要因に関する第１データ、及び前記対象装置への指令値であって、前記第１データにより示される前記要因に適応した指令値に関する第２データの組み合わせによりそれぞれ構成された複数件の学習用データセットを取得する学習データ取得部と、
    　取得した前記複数件の学習用データセットそれぞれについて、前記第１データを入力すると、前記第２データに対応する値を出力するように予測モデルを構築する学習処理部と、
    　取得した前記複数件の学習用データセットにおける前記第２データの分布から前記指令値の取り得る数値範囲を推定する推定部と、
    　前記対象装置への前記指令値に対して予め設定された第１の閾値により規定される第１の許容範囲を拡げるように、推定した前記数値範囲に基づいて、前記対象装置への前記指令値に対する第２の閾値を決定する閾値決定部と、
    　運用フェーズにおいて、前記要因に関する入力データを取得する入力データ取得部と、
    　取得した入力データを前記予測モデルに入力することで、前記予測モデルから出力値を取得し、取得した前記出力値に基づいて、決定した前記第２の閾値により規定される第２の許容範囲内で、前記対象装置に対する指令値を決定する予測演算部と、
    　決定した前記指令値に基づいて、前記対象装置の動作を制御する動作制御部と、
    を備える、
    制御システム。     First data relating to factors determining the operation of the target device, and command values to the target device, each comprising a combination of second data relating to command values adapted to the factors indicated by the first data A learning data acquisition unit for acquiring a plurality of learning data sets;
    For each of the acquired plurality of learning data sets, when the first data is input, a learning processing unit that builds a prediction model so as to output a value corresponding to the second data;
    An estimation unit that estimates a numerical range that the command value can take from the distribution of the second data in the acquired plurality of learning data sets;
    The command value to the target device based on the estimated numerical range so as to expand a first allowable range defined by a first threshold value set in advance for the command value to the target device. A threshold value determination unit for determining a second threshold value for
    In the operation phase, an input data acquisition unit that acquires input data related to the factors;
    By inputting the acquired input data to the prediction model, an output value is acquired from the prediction model, and the second allowable range defined by the second threshold determined based on the acquired output value And a prediction calculation unit for determining a command value for the target device,
    An operation control unit that controls the operation of the target device based on the determined command value;
    Comprising
    Control system.
  2. 　前記閾値決定部は、推定した前記数値範囲の境界値又は前記第１の閾値と当該境界値との間の値を前記第２の閾値として採用する、
    請求項１に記載の制御システム。     The threshold value determination unit employs the estimated boundary value of the numerical range or a value between the first threshold value and the boundary value as the second threshold value.
    The control system according to claim 1.
  3. 　前記第１の閾値は、前記第１の許容範囲の上限値であり、
    　前記閾値決定部は、前記上限値を超える値を前記第２の閾値として採用する、
    請求項１又は２に記載の制御システム。     The first threshold is an upper limit value of the first allowable range;
    The threshold value determination unit employs a value exceeding the upper limit value as the second threshold value.
    The control system according to claim 1 or 2.
  4. 　前記第１の閾値は、前記第１の許容範囲の下限値であり、
    　前記閾値決定部は、前記下限値より小さい値を前記第２の閾値として採用する、
    請求項１又は２に記載の制御システム。     The first threshold is a lower limit value of the first allowable range;
    The threshold value determination unit employs a value smaller than the lower limit value as the second threshold value.
    The control system according to claim 1 or 2.
  5. 　前記閾値決定部は、予め設定された安全条件を満たすように前記第２の閾値を決定する、
    請求項１から４のいずれか１項に記載の制御システム。     The threshold determination unit determines the second threshold so as to satisfy a preset safety condition;
    The control system according to any one of claims 1 to 4.
  6. 　前記第２データは、前記指令値の基準値に対する補正値により構成される、
    請求項１から５のいずれか１項に記載の制御システム。     The second data is constituted by a correction value with respect to a reference value of the command value.
    The control system according to any one of claims 1 to 5.
  7. 　前記対象装置は、ワークから製品を生産する生産装置であって、
    　前記第１データ及び前記入力データはそれぞれ、前記ワークの特徴量及び前記製品を生産する環境の属性値の少なくとも一方により構成される、
    請求項１から６のいずれか１項に記載の制御システム。     The target device is a production device for producing a product from a workpiece,
    Each of the first data and the input data is configured by at least one of a feature amount of the workpiece and an attribute value of an environment for producing the product.
    The control system according to any one of claims 1 to 6.
  8. 　コンピュータが、
    　対象装置の動作を決定する要因に関する第１データ、及び前記対象装置への指令値であって、前記第１データにより示される前記要因に適応した指令値に関する第２データの組み合わせによりそれぞれ構成された複数件の学習用データセットを取得するステップと、
    　取得した前記複数件の学習用データセットそれぞれについて、前記第１データを入力すると、前記第２データに対応する値を出力するように予測モデルを構築するステップと、
    　取得した前記複数件の学習用データセットにおける前記第２データの分布から前記指令値の取り得る数値範囲を推定するステップと、
    　前記対象装置への前記指令値に対して予め設定された第１の閾値により規定される第１の許容範囲を拡げるように、推定した前記数値範囲に基づいて、前記対象装置への前記指令値に対する第２の閾値を決定するステップと、
    　運用フェーズにおいて、前記要因に関する入力データを取得するステップと、
    　取得した入力データを前記予測モデルに入力することで、前記予測モデルから出力値を取得するステップと、
    　取得した前記出力値に基づいて、決定した前記第２の閾値により規定される第２の許容範囲内で、前記対象装置に対する指令値を決定するステップと、
    　決定した前記指令値に基づいて、前記対象装置の動作を制御するステップと、
    を実行する、
    制御方法。     Computer
    The first data relating to the factor that determines the operation of the target device and the command value to the target device, each comprising a combination of second data relating to the command value adapted to the factor indicated by the first data Acquiring a plurality of training data sets;
    For each of the acquired plurality of learning data sets, when the first data is input, constructing a prediction model so as to output a value corresponding to the second data;
    Estimating a possible numerical range of the command value from the distribution of the second data in the acquired plurality of learning data sets;
    The command value to the target device based on the estimated numerical range so as to expand a first allowable range defined by a first threshold value set in advance for the command value to the target device. Determining a second threshold for
    In the operational phase, obtaining input data relating to the factors;
    Obtaining the output value from the prediction model by inputting the acquired input data to the prediction model;
    Determining a command value for the target device within a second allowable range defined by the determined second threshold based on the acquired output value;
    Controlling the operation of the target device based on the determined command value;
    Run the
    Control method.
  9. 　対象装置の動作を決定する要因に関する第１データ、及び前記対象装置への指令値であって、前記第１データにより示される前記要因に適応した指令値に関する第２データの組み合わせによりそれぞれ構成された複数件の学習用データセットを取得する学習データ取得部と、
    　取得した前記複数件の学習用データセットそれぞれについて、前記第１データを入力すると、前記第２データに対応する値を出力するように予測モデルを構築する学習処理部と、
    　取得した前記複数件の学習用データセットにおける前記第２データの分布から前記指令値の取り得る数値範囲を推定する推定部と、
    　前記対象装置への前記指令値に対して予め設定された第１の閾値により規定される第１の許容範囲を拡げるように、推定した前記数値範囲に基づいて、前記対象装置への前記指令値に対する第２の閾値を決定する閾値決定部と、
    を備える、
    学習装置。     The first data relating to the factor that determines the operation of the target device and the command value to the target device, each comprising a combination of second data relating to the command value adapted to the factor indicated by the first data A learning data acquisition unit for acquiring a plurality of learning data sets;
    For each of the acquired plurality of learning data sets, when the first data is input, a learning processing unit that builds a prediction model so as to output a value corresponding to the second data;
    An estimation unit that estimates a numerical range that the command value can take from the distribution of the second data in the acquired plurality of learning data sets;
    The command value to the target device based on the estimated numerical range so as to expand a first allowable range defined by a first threshold value set in advance for the command value to the target device. A threshold value determining unit for determining a second threshold value for
    Comprising
    Learning device.
  10. 　対象装置の動作を決定する要因に関する入力データを取得する入力データ取得部と、
    　取得した入力データを前記予測モデルに入力することで、前記予測モデルから出力値を取得し、取得した前記出力値に基づいて、請求項９に記載の学習装置によって決定された前記第２の閾値により規定される第２の許容範囲内で、前記対象装置に対する指令値を決定する予測演算部と、
    　決定した前記指令値に基づいて、前記対象装置の動作を制御する動作制御部と、
    を備える、
    制御装置。     An input data acquisition unit that acquires input data relating to factors that determine the operation of the target device;
    The second threshold value determined by the learning device according to claim 9, wherein an input value is acquired from the prediction model by inputting the acquired input data to the prediction model, and based on the acquired output value. A prediction calculation unit that determines a command value for the target device within a second allowable range defined by
    An operation control unit that controls the operation of the target device based on the determined command value;
    Comprising
    Control device.
  11. 　コンピュータが、
    　対象装置の動作を決定する要因に関する第１データ、及び前記対象装置への指令値であって、前記第１データにより示される前記要因に適応した指令値に関する第２データの組み合わせによりそれぞれ構成された複数件の学習用データセットを取得するステップと、
    　取得した前記複数件の学習用データセットそれぞれについて、前記第１データを入力すると、前記第２データに対応する値を出力するように予測モデルを構築するステップと、
    　取得した前記複数件の学習用データセットにおける前記第２データの分布から前記指令値の取り得る数値範囲を推定するステップと、
    　前記対象装置への前記指令値に対して予め設定された第１の閾値により規定される第１の許容範囲を拡げるように、推定した前記数値範囲に基づいて、前記対象装置への前記指令値に対する第２の閾値を決定するステップと、
    を実行する、
    学習方法。     Computer
    The first data relating to the factor that determines the operation of the target device and the command value to the target device, each comprising a combination of second data relating to the command value adapted to the factor indicated by the first data Acquiring a plurality of training data sets;
    For each of the acquired plurality of learning data sets, when the first data is input, constructing a prediction model so as to output a value corresponding to the second data;
    Estimating a possible numerical range of the command value from the distribution of the second data in the acquired plurality of learning data sets;
    The command value to the target device based on the estimated numerical range so as to expand a first allowable range defined by a first threshold value set in advance for the command value to the target device. Determining a second threshold for
    Run the
    Learning method.
  12. 　コンピュータに、
    　対象装置の動作を決定する要因に関する第１データ、及び前記対象装置への指令値であって、前記第１データにより示される前記要因に適応した指令値に関する第２データの組み合わせによりそれぞれ構成された複数件の学習用データセットを取得するステップと、
    　取得した前記複数件の学習用データセットそれぞれについて、前記第１データを入力すると、前記第２データに対応する値を出力するように予測モデルを構築するステップと、
    　取得した前記複数件の学習用データセットにおける前記第２データの分布から前記指令値の取り得る数値範囲を推定するステップと、
    　前記対象装置への前記指令値に対して予め設定された第１の閾値により規定される第１の許容範囲を拡げるように、推定した前記数値範囲に基づいて、前記対象装置への前記指令値に対する第２の閾値を決定するステップと、
    を実行させるための、
    学習プログラム。     On the computer,
    The first data relating to the factor that determines the operation of the target device and the command value to the target device, each comprising a combination of second data relating to the command value adapted to the factor indicated by the first data Acquiring a plurality of training data sets;
    For each of the acquired plurality of learning data sets, when the first data is input, constructing a prediction model so as to output a value corresponding to the second data;
    Estimating a possible numerical range of the command value from the distribution of the second data in the acquired plurality of learning data sets;
    The command value to the target device based on the estimated numerical range so as to expand a first allowable range defined by a first threshold value set in advance for the command value to the target device. Determining a second threshold for
    To run
    Learning program.
PCT/JP2019/010189 2018-06-07 2019-03-13 Control system, control method, learning device, control device, learning method, and learning program WO2019235018A1 (en)

Priority Applications (3)

Application Number Priority Date Filing Date Title
US17/052,523 US11681261B2 (en) 2018-06-07 2019-03-13 Control system, control method, learning device, control device, learning method for controlling an operation of a subject device on basis of a detemined command value
CN201980030426.2A CN112105994B (en) 2018-06-07 2019-03-13 Control system, apparatus and method, learning apparatus and method, and medium
EP19814529.4A EP3805875B1 (en) 2018-06-07 2019-03-13 Control system and method for controlling the operation of a production device

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
JP2018109630A JP7031502B2 (en) 2018-06-07 2018-06-07 Control system, control method, learning device, control device, learning method and learning program
JP2018-109630 2018-06-07

Publications (1)

Publication Number Publication Date
WO2019235018A1 true WO2019235018A1 (en) 2019-12-12

Family

ID=68769305

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/JP2019/010189 WO2019235018A1 (en) 2018-06-07 2019-03-13 Control system, control method, learning device, control device, learning method, and learning program

Country Status (6)

Country Link
US (1) US11681261B2 (en)
EP (1) EP3805875B1 (en)
JP (1) JP7031502B2 (en)
CN (1) CN112105994B (en)
TW (1) TW202001706A (en)
WO (1) WO2019235018A1 (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US11138003B2 (en) * 2019-04-02 2021-10-05 Taplytics Inc. Methods and systems for automatic determination of a device-specific configuration for a software application operating on a user device

Families Citing this family (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP7031502B2 (en) * 2018-06-07 2022-03-08 オムロン株式会社 Control system, control method, learning device, control device, learning method and learning program
JP2021096639A (en) * 2019-12-17 2021-06-24 キヤノン株式会社 Control method, controller, mechanical equipment, control program, and storage medium
JP7299184B2 (en) * 2020-03-11 2023-06-27 日立Ｇｅニュークリア・エナジー株式会社 HYDRAULIC DRIVE AND METHOD OF CONTROLLING HYDRAULIC DRIVE
JP2022127675A (en) * 2021-02-22 2022-09-01 オムロン株式会社 Model generation device, model generation method, model generation program, estimation device, estimation method and estimation program

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPH0863203A (en) * 1994-08-24 1996-03-08 Hitachi Ltd Method and device for neutral net control
JPH0988685A (en) * 1995-09-29 1997-03-31 Matsushita Electric Ind Co Ltd Air-fuel ratio control device for internal combustion engine
JP2000099107A (en) 1998-09-25 2000-04-07 Hitachi Ltd Model predictive control system
JP2017102613A (en) * 2015-11-30 2017-06-08 ファナック株式会社 Machine learning device and method for optimizing smoothness of feeding of feed shaft of machine and motor control device having machine learning device

Family Cites Families (15)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20060247798A1 (en) 2005-04-28 2006-11-02 Subbu Rajesh V Method and system for performing multi-objective predictive modeling, monitoring, and update for an asset
US7526463B2 (en) 2005-05-13 2009-04-28 Rockwell Automation Technologies, Inc. Neural network using spatially dependent data for controlling a web-based process
KR101096793B1 (en) * 2010-04-19 2011-12-21 비앤에프테크놀로지 주식회사 data collecting method for detection and on-time warning system of industrial process
US10318882B2 (en) * 2014-09-11 2019-06-11 Amazon Technologies, Inc. Optimized training of linear machine learning models
US10339465B2 (en) * 2014-06-30 2019-07-02 Amazon Technologies, Inc. Optimized decision tree based models
US11086310B2 (en) * 2015-05-27 2021-08-10 Honeywell International Inc. Method and apparatus for real time model predictive control operator support in industrial process control and automation systems
US9904916B2 (en) * 2015-07-01 2018-02-27 Klarna Ab Incremental login and authentication to user portal without username/password
US10732621B2 (en) * 2016-05-09 2020-08-04 Strong Force Iot Portfolio 2016, Llc Methods and systems for process adaptation in an internet of things downstream oil and gas environment
CN107290965B (en) 2017-08-01 2019-11-08 浙江大学 Adaptive soft-sensor prediction technique based on local weighted Bayesian network
JP6797780B2 (en) * 2017-11-07 2020-12-09 株式会社東芝 Information processing equipment, information processing methods and programs
JP7019396B2 (en) * 2017-11-30 2022-02-15 日本電産マシンツール株式会社 Machine tool control method, machine tool control device, machine tool setting support device, machine tool control system and program
CN108076224B (en) 2017-12-21 2021-06-29 Oppo广东移动通信有限公司 Application program control method and device, storage medium and mobile terminal
JP6888577B2 (en) * 2018-03-30 2021-06-16 オムロン株式会社 Control device, control method, and control program
JP6740278B2 (en) * 2018-04-13 2020-08-12 ファナック株式会社 Machine learning device, control device, and machine learning method
JP7031502B2 (en) * 2018-06-07 2022-03-08 オムロン株式会社 Control system, control method, learning device, control device, learning method and learning program

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPH0863203A (en) * 1994-08-24 1996-03-08 Hitachi Ltd Method and device for neutral net control
JPH0988685A (en) * 1995-09-29 1997-03-31 Matsushita Electric Ind Co Ltd Air-fuel ratio control device for internal combustion engine
JP2000099107A (en) 1998-09-25 2000-04-07 Hitachi Ltd Model predictive control system
JP2017102613A (en) * 2015-11-30 2017-06-08 ファナック株式会社 Machine learning device and method for optimizing smoothness of feeding of feed shaft of machine and motor control device having machine learning device

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US11138003B2 (en) * 2019-04-02 2021-10-05 Taplytics Inc. Methods and systems for automatic determination of a device-specific configuration for a software application operating on a user device

Also Published As

Publication number Publication date
EP3805875A1 (en) 2021-04-14
JP7031502B2 (en) 2022-03-08
EP3805875A4 (en) 2022-03-02
TW202001706A (en) 2020-01-01
EP3805875B1 (en) 2023-10-18
CN112105994A (en) 2020-12-18
US20210240144A1 (en) 2021-08-05
JP2019212164A (en) 2019-12-12
US11681261B2 (en) 2023-06-20
CN112105994B (en) 2022-10-11

Similar Documents

Publication Publication Date Title
WO2019235018A1 (en) Control system, control method, learning device, control device, learning method, and learning program
US10121107B2 (en) Machine learning device and method for optimizing frequency of tool compensation of machine tool, and machine tool having the machine learning device
JP6888577B2 (en) Control device, control method, and control program
US10935967B2 (en) Machining equipment system and manufacturing system
EP3405843B1 (en) Model adaption and online learning for unstable environments
US10442023B2 (en) Simulation apparatus of wire electric discharge machine having function of determining welding positions of core using machine learning
JP2017033138A (en) Machine tool, simulation apparatus, and machine learning device
JP7151547B2 (en) Predictive control development device, predictive control development method, and predictive control development program
US11762344B2 (en) Control device, control method, and control program
WO2016092872A1 (en) Control device, program therefor, and plant control method
GB2593604A (en) Prediction model enhancement
US11579000B2 (en) Measurement operation parameter adjustment apparatus, machine learning device, and system
JP6958461B2 (en) Control device, control method, and control program
JP4524683B2 (en) Parameter adjustment device for plant model
JP6659260B2 (en) Control device and machine learning device with PLC program optimization function
JP6909616B2 (en) Manufacturing system and manufacturing method
JP2020071510A (en) Production system, production system control method, and production control device
WO2022210472A1 (en) Machining condition adjustment device

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 19814529

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

ENP Entry into the national phase

Ref document number: 2019814529

Country of ref document: EP

Effective date: 20210111